Security Operations Centers have spent years improving their ability to see what is happening across the enterprise. Modern security teams collect signals from endpoints, networks, cloud infrastructure, identity platforms, applications, email systems, firewalls, vulnerability tools, SIEM platforms, EDR solutions, and an expanding security stack. In theory, greater visibility should make organizations easier to defend. In practice, it has created another problem: security teams can generate and collect information faster than people can investigate it.
Every day, a SOC may receive an enormous volume of alerts ranging from routine anomalies and known false positives to activity that could indicate a genuine compromise. Each alert potentially requires someone to establish what happened, determine whether the behavior is suspicious, gather additional context, assess its severity, document the investigation, and decide what should happen next. Most alerts will not become major incidents, but the analyst often has to do enough work to establish that before the alert can safely be closed.
The traditional SOC model manages this workload through tiers. Tier 1 analysts provide the first line of investigation, monitoring alerts, performing initial triage, gathering context, and escalating suspicious activity. Tier 2 analysts investigate incidents requiring deeper analysis, correlate evidence across systems, and determine the scope and severity of potential threats. Tier 3 analysts handle the most complex investigations, advanced threats, threat hunting, detection engineering, and incidents requiring significant technical expertise.
The structure is logical because it attempts to reserve scarce expertise for the problems that genuinely require it. A senior security analyst should not spend the majority of the day investigating routine alerts that could have been resolved during initial triage. Tier 1 should filter and enrich the workload before it moves upward, Tier 2 should conduct deeper investigation, and Tier 3 should concentrate on the threats where advanced expertise creates the greatest value.
The problem appears when alert volume grows faster than the SOC's investigative capacity.
Tier 1 analysts can become overwhelmed by repetitive triage. Alerts begin accumulating in queues. Tier 2 receives escalations without enough context and has to repeat investigative work. Experienced analysts get pulled into incidents that could have been resolved earlier. Information needed to understand one alert may be distributed across the SIEM, EDR, identity platform, threat intelligence feeds, cloud logs, and ticketing system, forcing analysts to move between tools simply to reconstruct what happened.
This is not only an efficiency problem. It can affect the quality of security operations. When analysts are processing large numbers of repetitive alerts under time pressure, attention becomes a scarce resource. Genuine threats have to compete with false positives and low-risk activity for investigative capacity. Meanwhile, attackers do not wait for the SOC queue to clear before continuing reconnaissance, escalating privileges, moving laterally, or attempting to achieve their objective.
Artificial intelligence is beginning to change this operating model.
AI can assist with many of the activities that consume analyst time before and during an investigation. It can gather context from multiple security systems, correlate related signals, summarize alert histories, interpret logs, enrich indicators, reconstruct timelines, surface relevant threat intelligence, recommend investigative steps, draft incident documentation, and automate repetitive response workflows. Instead of analysts manually assembling every piece of context, AI can increasingly help prepare the investigation for them.
The important implication is not necessarily that Tier 1, Tier 2, or Tier 3 analysts disappear. It is that the division of work between humans and machines begins to change at every tier.
A Tier 1 analyst supported by AI may spend less time gathering basic context and more time validating conclusions and identifying unusual behavior. A Tier 2 analyst may begin an investigation with a correlated timeline and summarized evidence rather than a minimally enriched alert. A Tier 3 analyst may use AI to accelerate threat hunting, analyze larger datasets, develop detections, and investigate sophisticated activity without spending as much time on repetitive investigative preparation.
This creates a more useful way to think about AI in security operations. The question is not simply whether AI can perform tasks currently handled by SOC analysts. Many individual tasks increasingly can be automated or accelerated.
The more consequential question is:
What happens to the SOC operating model when every Tier 1, Tier 2, and Tier 3 analyst has AI working alongside them?
That is where the next stage of security operations becomes much more interesting.
What Are Tier 1, Tier 2, and Tier 3 SOC Analysts?
Before looking at how AI is changing security operations, it helps to understand why Security Operations Centers developed tiered analyst structures in the first place. A SOC can receive alerts ranging from obvious false positives to sophisticated attacks that require hours or days of investigation. Giving every event to the most experienced security professional would be inefficient and prohibitively expensive, so many organizations divide investigative work according to complexity, risk, and the level of expertise required.
The exact responsibilities vary between organizations, and not every SOC uses formal Tier 1, Tier 2, and Tier 3 titles. Smaller security teams may combine several responsibilities into a single role, while larger enterprises and managed security providers may use more specialized structures. But the underlying principle is common: routine alerts are handled first, suspicious activity receives deeper investigation, and the most complex threats are escalated to people with more specialized expertise.
i. Tier 1 SOC Analyst: Alert Monitoring and Initial Triage
Tier 1 analysts typically sit at the front of the security operations process. They monitor incoming alerts, review security events, perform initial triage, gather basic context, check indicators, document findings, and determine whether an event can be closed or requires further investigation. In many environments, they are responsible for processing a significant proportion of the SOC's overall alert volume.
The central question at this stage is relatively straightforward:
Is this alert something we need to investigate further?
Answering it, however, can require more work than the question suggests. An analyst may need to identify the affected user or endpoint, review recent activity, examine the alert history, check an IP address or domain against threat intelligence, determine whether similar events have occurred elsewhere, and establish whether the behavior is expected within the organization's environment. Only after gathering that context can the analyst make a reasonable decision about whether to close, investigate, or escalate the alert.
A large part of Tier 1 work therefore involves separating genuine security concerns from false positives, duplicate alerts, known activity, and events that do not justify deeper investigation. The role is essential because every alert that can be confidently resolved at this stage prevents unnecessary investigative work further up the SOC.
It is also one of the areas where repetitive workload can become particularly significant. When analysts repeatedly perform the same enrichment, lookup, documentation, and triage steps across large numbers of alerts, considerable human capacity is consumed before a genuine incident ever reaches deeper investigation.
ii. Tier 2 SOC Analyst: Investigation and Incident Analysis
Tier 2 analysts generally receive alerts or incidents that could not be confidently resolved during initial triage. At this point, the question changes. The objective is no longer simply to determine whether an alert looks suspicious. The analyst needs to understand what actually happened, how serious it is, what may have been affected, and what the organization should do next.
That can require correlating information across several systems. A Tier 2 analyst might examine endpoint telemetry, authentication events, network activity, process execution, cloud logs, user behavior, email activity, threat intelligence, and other evidence to reconstruct an incident. An unusual login, for example, may need to be examined alongside device information, geographic anomalies, subsequent account activity, privilege changes, mailbox behavior, and activity on other systems before its significance becomes clear.
Tier 2 analysts may also determine the scope of an incident. If one endpoint appears compromised, is the activity isolated to that device or visible elsewhere? If credentials may have been stolen, were they subsequently used to access other systems? If malware executed, what processes did it create, what network connections followed, and did the attacker establish persistence?
Depending on the organization's operating model, Tier 2 may also coordinate or initiate containment and remediation actions. That could include isolating an endpoint, disabling an account, blocking an indicator, requesting credential resets, coordinating with IT teams, or escalating the incident further when more advanced investigation is required.
The quality of Tier 1 triage has a direct effect on this work. If an incident reaches Tier 2 with incomplete context, the analyst may have to repeat many of the same investigative steps before deeper analysis can begin. This is one of the inefficiencies AI has the potential to reduce.
iii. Tier 3 SOC Analyst: Advanced Investigation and Threat Hunting
Tier 3 analysts typically handle the SOC's most complex technical problems. By the time an incident reaches this level, the objective may extend well beyond determining whether an individual alert is malicious. Analysts may need to understand an attacker's techniques, reconstruct a broader intrusion, determine the full scope of compromise, identify persistence mechanisms, analyze malicious code, or discover activity that existing security controls failed to detect.
Responsibilities can include advanced incident investigation, threat hunting, malware analysis, digital forensics, detection engineering, attack-path analysis, and investigation of sophisticated or previously unseen threats. Tier 3 analysts may also work closely with incident response, security engineering, threat intelligence, cloud security, identity teams, and other specialists depending on the nature of the incident.
Threat hunting is an important distinction at this level. Much of Tier 1 and Tier 2 work begins because a security system has generated an alert. Threat hunting can reverse that relationship. Instead of waiting for a detection to fire, analysts develop hypotheses about how an attacker might operate and search telemetry for evidence of activity that existing rules or controls may have missed.
For example, a Tier 3 analyst investigating a compromised identity may not stop after identifying the original malicious login. They may examine whether the attacker changed authentication methods, created persistence, accessed cloud resources, elevated privileges, moved laterally, interacted with sensitive systems, or used techniques that could remain undetected elsewhere in the environment. The objective becomes understanding the attack as a whole rather than resolving an individual alert.
Because these skills are more specialized and relatively scarce, organizations generally want Tier 3 analysts concentrating on problems that genuinely require them. Every hour a senior analyst spends reconstructing information that could have been collected earlier is an hour unavailable for advanced investigation, threat hunting, detection improvement, or other high-value security work.
Why SOC Tiers Exist
The traditional SOC tier structure exists for a practical reason: not every security event deserves the same amount of human expertise. A routine alert that can be resolved in a few minutes should not consume the time of a senior incident responder, while a sophisticated intrusion should not remain stuck in an initial triage queue. Dividing work across Tier 1, Tier 2, and Tier 3 allows organizations to match investigative effort with the complexity and potential severity of the security problem.
At its simplest, the escalation path looks like this:
Tier 1: Is this alert worth investigating?
Tier 2: What happened, how serious is it, and what should we do?
Tier 3: What is the full scope of the threat, how did the attacker operate, and what might we still be missing?
The structure is therefore about more than seniority. It is a mechanism for managing scarce investigative capacity. High-volume and relatively predictable work remains at Tier 1 whenever possible. Incidents requiring deeper correlation and analysis move to Tier 2. The most technically complex investigations reach Tier 3, where specialized expertise can be concentrated on threat hunting, advanced incident response, detection engineering, forensics, and sophisticated attacks.
The weakness appears when the volume of work entering the SOC exceeds the capacity available at one or more of those levels. Tier 1 analysts become occupied with repetitive triage and enrichment. Alerts accumulate before they can be investigated. Tier 2 receives too many escalations, sometimes without enough context to begin meaningful analysis, forcing analysts to repeat work that should have happened earlier. Tier 3 can then become involved in incidents that consume senior expertise without necessarily requiring it.
In that environment, the tier structure can unintentionally become a series of queues. An alert waits for Tier 1 investigation, an escalation waits for Tier 2 analysis, and a complex incident waits for Tier 3 expertise. Each handoff can introduce additional time, documentation, and context gathering. The model that was designed to protect scarce security expertise can begin consuming that expertise simply through the mechanics of moving investigations between people.
AI changes this equation because some of the work that traditionally justified those handoffs can increasingly happen automatically. Alert enrichment, indicator lookups, event correlation, historical searches, timeline construction, incident summarization, documentation, and predefined investigative steps can potentially occur before an analyst begins working on the case. Instead of receiving a raw alert and manually assembling its context, the analyst can begin with a much richer investigative picture.
That does not necessarily make Tier 1, Tier 2, or Tier 3 obsolete. It changes where human expertise enters the investigation and what the analyst is expected to contribute once it does. Tier 1 can move away from repetitive information gathering toward validation and judgment. Tier 2 can spend less time reconstructing basic context and more time determining scope, impact, and response. Tier 3 can concentrate more heavily on advanced threats, novel attack behavior, threat hunting, and improving the organization's detections.
The traditional question has been:
➡️ Which analyst tier should receive this alert next?
In an AI-assisted SOC, another question becomes equally important:
➡️ How much investigation can be completed before the next analyst needs to become involved at all?
That shift from routing work between analysts to preparing work for analysts is one of the most important ways AI could reshape the traditional SOC operating model.
How AI Changes the Traditional SOC Workflow
Traditional SOC investigations are heavily dependent on analysts gathering, connecting, and interpreting information. An alert may originate in one security platform, but understanding whether it represents a genuine threat often requires evidence from several others. Before an analyst can make a decision, they first have to reconstruct enough context to understand what actually happened.
Consider a relatively typical alert. An analyst opens it and identifies the affected endpoint or account. They review the activity that triggered the detection, search relevant logs, examine recent authentication events, check the user's behavior, investigate an IP address, domain, file hash, or process, and look for related activity elsewhere in the environment. They may then consult threat intelligence, compare the event with previous incidents, document their findings, and determine whether the alert can be closed, requires further investigation, or should be escalated.
The individual steps are necessary. The problem is how much analyst time can be consumed simply assembling the evidence required to begin making a security decision.
This becomes particularly inefficient when the information is distributed across multiple systems. An analyst might move between a SIEM, EDR platform, identity provider, cloud security tool, threat intelligence service, and ticketing system during a single investigation. Even when each lookup takes only a few minutes, the cumulative workload becomes substantial when the same process is repeated across hundreds or thousands of alerts.
AI has the potential to compress parts of this investigative process.
Instead of waiting for an analyst to manually perform every lookup, an AI-assisted security workflow can potentially begin gathering relevant context as soon as an alert is generated. It can retrieve information about the affected asset and user, examine related events, enrich indicators with available threat intelligence, identify previous alerts involving the same entities, correlate activity across connected systems, and organize the evidence into a more usable investigative picture.
The analyst can then begin with something closer to:
What happened → Which users and systems are involved → What supporting evidence exists → What related activity was found → Why the activity may matter → What should be investigated next
That is a fundamentally different starting point from receiving a raw alert.
For example, instead of seeing only an alert for a suspicious authentication event, an analyst might receive context showing that the account logged in from an unusual location, the authentication came from a previously unseen device, several failed attempts occurred beforehand, a privileged application was accessed shortly afterward, and the source IP has relevant threat intelligence associated with it. The system might also identify that no similar activity has been observed for that user during the previous 30 days and recommend several additional checks before the incident is closed or escalated.
The AI is not necessarily making the final security decision. It is reducing the investigative distance between the alert and the decision.
This distinction matters. Security investigations involve ambiguity, incomplete evidence, changing attacker behavior, and organizational context that may not be visible to an automated system. An unusual login may represent account compromise, but it may also be legitimate travel. A suspicious PowerShell command could indicate malicious execution or an approved administrative process. A connection to an unfamiliar domain may be malicious, or it may belong to a newly deployed business application. Context still needs interpretation.
The role of the analyst therefore shifts rather than simply disappearing. Instead of spending as much time retrieving information, analysts can spend more time evaluating evidence, challenging assumptions, investigating anomalies, determining impact, and deciding what action is appropriate. AI handles more of the mechanical work surrounding an investigation while human expertise remains responsible for the judgments that carry security consequences.
There is also an important difference between assistance and autonomy. An AI system that summarizes evidence for an analyst carries a different level of operational risk from one that automatically disables an account, isolates an endpoint, blocks network traffic, or closes an incident. As the consequences of an automated action increase, organizations need stronger controls around confidence, permissions, validation, auditability, and human approval.
The goal, therefore, should not simply be to remove analysts from investigations. It should be to determine how much repetitive investigative work can be completed before human judgment becomes necessary.
And that answer looks different across the SOC.
For Tier 1, AI can substantially change alert enrichment and triage. For Tier 2, it can accelerate correlation, investigation, and incident analysis. For Tier 3, it can expand the amount of data analysts can examine during advanced investigations and threat hunting.
AI affects all three tiers, but it does not affect them in the same way.
How AI Transforms Tier 1 SOC Analysts
Tier 1 is arguably where AI can create the most immediate operational impact inside a Security Operations Center. The reason is not that Tier 1 work is simple. It is that a significant portion of the workload is high-volume, repetitive, and structured enough for AI-assisted investigation.
A Tier 1 analyst may process dozens of alerts during a shift, and many require a similar sequence of actions before any meaningful decision can be made. Identify the affected user or asset. Retrieve relevant logs. Check the source IP. Review authentication activity. Look for related alerts. Consult threat intelligence. Determine whether the activity fits expected behavior. Document the findings. Then decide whether the alert can be closed or needs to move deeper into the SOC.
AI has the potential to perform or accelerate many of those preparatory steps. That changes Tier 1 from a role heavily focused on collecting information into one increasingly focused on evaluating information and deciding what deserves human attention next.
i. From Alert Review to AI-Assisted Triage
Consider a suspicious login alert. Traditionally, the alert itself may provide only part of the information an analyst needs. Before deciding whether the activity represents an actual security concern, the Tier 1 analyst may need to establish:
- which user triggered the alert
- where the authentication originated
- whether the device is recognized
- whether the source IP has a suspicious reputation
- whether similar authentication attempts occurred recently
- whether MFA was successfully completed
- whether other alerts involve the same account or endpoint
- whether the behavior differs from the user's normal activity
- whether the event meets the organization's escalation criteria
None of these steps is individually extraordinary. The operational problem is that analysts may need to repeat variations of them across large numbers of alerts every day.
An AI-assisted workflow can potentially begin collecting this information as soon as the alert appears. It can retrieve relevant identity and endpoint information, examine recent authentication history, enrich IP addresses or domains using available threat intelligence, search for related alerts, and organize the results into a concise summary. Instead of beginning with an isolated security signal, the analyst begins with a more complete picture of the event and the surrounding activity.
For example, rather than receiving only a "suspicious login" alert, the analyst could receive a summary indicating that the account authenticated from a previously unseen device, the source IP has relevant threat intelligence, several unsuccessful login attempts occurred immediately beforehand, MFA was completed, and unusual access to a sensitive application followed shortly afterward. The system could also surface related endpoint or identity alerts and identify which additional checks are required by the organization's investigation playbook.
The analyst's job then changes. Less time is spent discovering basic facts, and more time can be spent asking whether the assembled evidence makes sense, whether important context is missing, and whether the incident warrants escalation.
ii. Reducing Repetitive Alert Triage
One of the persistent challenges in security operations is that individual alerts do not always communicate risk particularly well. A single event may appear relatively harmless when viewed in isolation but become much more significant when connected with other activity involving the same user, endpoint, application, or infrastructure.
An isolated failed login, for example, may have little significance. Failed authentications happen constantly in most environments. But repeated failed attempts followed by a successful authentication from an unusual location, a change in account privileges, and suspicious endpoint activity present a very different security picture.
AI can help connect these signals by examining relationships between events rather than requiring analysts to evaluate every alert independently. It can potentially identify shared entities, related timelines, unusual sequences of behavior, and historical patterns that make one alert more or less significant when viewed in context.
This does not mean AI should automatically determine that every correlated sequence represents an attack. Security environments contain legitimate anomalies, administrative activity, unusual employee behavior, and business processes that can resemble malicious activity. Human validation remains important.
The operational benefit is that Tier 1 analysts can spend less time repeatedly reconstructing relationships between alerts and more time examining the cases where the combined evidence suggests meaningful risk.
iii. Automating Initial Alert Enrichment
Alert enrichment is one of the clearest examples of work that AI and automation can move earlier in the investigation. Security analysts frequently need information from multiple systems before they can understand an alert, and retrieving that information manually can involve moving between numerous consoles and running the same searches repeatedly.
Depending on the organization's security stack, relevant context might come from:
- SIEM platforms
- EDR or XDR systems
- identity and access management platforms
- threat intelligence services
- asset inventories or CMDBs
- vulnerability management systems
- cloud security platforms
- email security tools
- network security systems
- historical alerts and incident records
The important change is not simply that AI can access more information. It is that the system can potentially determine which information is relevant to the alert and assemble it before the analyst begins investigating.
A suspicious process execution alert, for example, might automatically be enriched with the endpoint owner, asset criticality, vulnerability status, parent and child processes, file reputation, recent network connections, previous alerts involving the endpoint, and similar activity elsewhere in the environment. An analyst can then investigate the security meaning of those facts instead of spending the first several minutes retrieving them.
At scale, removing even a few minutes of manual enrichment from a large proportion of Tier 1 alerts can translate into substantial additional investigative capacity.
iv. Improving Alert Prioritization
AI can also potentially improve how alerts are prioritized before analysts begin working through the queue. Traditional severity ratings are often determined by individual detection rules or predetermined scoring systems. Those ratings may not fully account for the wider business and security context surrounding an event.
The same technical behavior can carry very different implications depending on where it occurs. Suspicious activity involving a low-value test device may warrant investigation, but the same behavior involving a privileged administrator account or a critical production server may deserve considerably faster attention.
AI-assisted prioritization can potentially combine the alert itself with contextual factors such as asset criticality, identity privileges, vulnerability exposure, threat intelligence, related detections, and historical behavior. This can help the SOC move toward prioritizing incidents in context rather than alerts in isolation.
Human oversight remains important because prioritization models can be wrong, incomplete, or influenced by poor underlying data. But even when AI does not make the final decision, better contextual prioritization can help Tier 1 analysts direct limited attention toward the events most likely to matter.
v. Improving Alert Documentation and Tier 2 Handoffs
Documentation is essential to effective security operations, but it also consumes analyst time. After completing triage, a Tier 1 analyst may need to record what was investigated, which evidence was reviewed, what conclusions were reached, and why an incident was closed or escalated.
AI can assist by generating a structured incident summary from the evidence collected during triage. Instead of writing every note manually, the analyst can review, correct, and approve a draft containing the relevant events, entities, investigative actions, findings, and escalation rationale.
The value becomes particularly apparent when an incident moves from Tier 1 to Tier 2.
A weak escalation might tell Tier 2 that an account showed suspicious activity and requires further investigation. A strong escalation can explain which user and systems are involved, what triggered the detection, which evidence Tier 1 reviewed, what unusual behavior was identified, which hypotheses have already been ruled out, and exactly why deeper investigation is required.
That reduces duplicated work between tiers. Tier 2 does not need to reconstruct the entire investigation simply to understand why Tier 1 escalated it.
In this sense, AI can improve not only the productivity of the individual analyst but also the flow of investigative context through the SOC.
The New Tier 1 SOC Analyst
AI does not necessarily eliminate Tier 1. But it can change what strong Tier 1 performance looks like.
Historically, a substantial part of entry-level SOC work has involved learning through repetition: examining alerts, performing lookups, following playbooks, gathering evidence, documenting findings, and escalating when predefined conditions are met. As more of that mechanical work becomes automated, the value of the analyst increasingly moves toward interpretation.
The Tier 1 analyst of an AI-assisted SOC may spend more time validating AI-generated findings, identifying missing context, recognizing unusual behavior, challenging incorrect conclusions, handling exceptions, assessing risk, and making better escalation decisions. Instead of asking analysts to become faster at moving between security consoles, organizations can increasingly ask them to become better at understanding what the evidence actually means.
That shift also has implications for how entry-level security professionals are trained. If AI performs more of the routine information gathering that historically gave junior analysts exposure to security tools and investigation patterns, SOC leaders will need to ensure analysts still develop the underlying knowledge required to recognize when the AI is wrong. An analyst cannot effectively validate an investigation they do not understand.
AI therefore raises rather than removes the importance of foundational security knowledge. Tier 1 analysts still need to understand authentication, endpoint behavior, network activity, common attack techniques, organizational context, and investigative methodology. The difference is that they may be expected to apply that knowledge to judgment and validation much earlier in their careers.
The biggest change to Tier 1 is therefore not simply faster alert processing.
It is a shift from manually assembling the investigation to critically evaluating an investigation that has already been partially assembled for them.
How AI Transforms Tier 2 SOC Analysts
If Tier 1 is primarily concerned with determining which alerts deserve further attention, Tier 2 is where the investigation becomes deeper. The analyst is no longer asking only whether something looks suspicious. They are trying to understand what happened, how the activity developed, which users and systems were affected, how serious the incident may be, and what the organization should do next.
This makes Tier 2 particularly well suited to AI-assisted investigation. A significant amount of analyst time can be spent searching across systems, reconstructing timelines, correlating evidence, testing hypotheses, and documenting findings before a confident conclusion can be reached. AI can accelerate many of those activities, allowing analysts to reach the parts of an investigation that require security judgment much sooner.
The opportunity is therefore not to hand the entire investigation to AI. It is to reduce the amount of manual investigative preparation between receiving an incident and understanding it.
i. Building Incident Timelines Faster
One of the most time-consuming parts of incident investigation is reconstructing the sequence of events. Security incidents rarely appear as a clean narrative. Evidence is distributed across authentication logs, endpoint telemetry, cloud activity, network events, email systems, security alerts, and other sources, often with different timestamps, identifiers, and levels of detail.
A Tier 2 analyst investigating a potentially compromised account, for example, might need to determine when the suspicious activity began, which device was involved, what happened immediately before and after authentication, which applications were accessed, whether privileges changed, what processes executed, and whether data or other systems were subsequently accessed.
The value is not simply presenting events chronologically. A useful AI-assisted system can help identify which events may be related, connect entities across different sources, remove irrelevant noise, and highlight gaps that require additional investigation.
Instead of spending the first part of the investigation manually constructing the sequence, the Tier 2 analyst receives a hypothesis about what happened and the evidence supporting it. The analyst can then verify the underlying events, challenge incorrect associations, investigate missing context, and determine whether the proposed timeline accurately represents the incident.
That distinction is important. AI can accelerate timeline construction, but the resulting narrative should not automatically be treated as fact. Correlation does not always establish causation, timestamps can be misleading, and legitimate activity can occur alongside malicious behavior. Human verification remains essential.
ii. Correlating Signals Across Security Tools
Attackers rarely organize their activity conveniently inside the boundaries of a single security product. A compromised identity may first appear as an unusual authentication event. The endpoint platform may later identify suspicious process execution. A cloud security tool may detect abnormal access to a storage resource. The SIEM may contain additional events, while threat intelligence provides context about infrastructure associated with the activity.
Viewed independently, none of those signals may provide enough evidence to establish that a serious incident is underway.
Viewed together, they can tell a very different story.
AI can help Tier 2 analysts identify these relationships across security systems by connecting users, devices, IP addresses, processes, domains, applications, cloud resources, and other entities. Instead of manually searching each platform for related activity, the analyst can begin with a broader view of the incident and then investigate the connections that appear most significant.
This becomes increasingly valuable as security stacks grow. Organizations have invested heavily in visibility, but greater visibility also means more places an analyst may need to search during an investigation. AI can help turn that fragmented visibility into usable context.
The operational shift is from asking analysts to find every relevant signal themselves toward asking them to evaluate the relationships between signals that have already been surfaced.
iii. Expanding the Scope of an Investigation
Once malicious activity is suspected, one of the most important Tier 2 responsibilities is determining scope. Finding one compromised account or endpoint is only the beginning. The analyst needs to establish whether the activity extends elsewhere in the environment.
AI can accelerate this process by searching for related indicators, behaviors, identities, devices, and infrastructure across available telemetry. If a malicious process is identified on one endpoint, the system can potentially look for the same or similar execution elsewhere. If an attacker used a particular IP address or domain, AI-assisted workflows can search historical activity involving that infrastructure. If an identity is compromised, the investigation can expand to applications, systems, sessions, and resources accessed by that account.
This can help answer questions such as:
- Did the same behavior occur on other endpoints?
- Were additional accounts targeted?
- Did the attacker access privileged systems?
- Was the same infrastructure observed elsewhere?
- Did the compromised identity create new sessions or credentials?
- Was there evidence of lateral movement?
- Did suspicious activity begin earlier than the original alert suggested?
- Are there indicators that the incident is still active?
The objective is not simply to process the alert that triggered the investigation. It is to determine whether that alert represents one visible part of a larger incident.
AI can make it easier to broaden that search without requiring the analyst to construct and execute every query manually. The Tier 2 analyst remains responsible for deciding which relationships are meaningful and when the evidence is sufficient to support a conclusion.
iv. Supporting Root-Cause Analysis
Understanding what happened is only part of incident analysis. Tier 2 also needs to investigate how it happened.
A compromised account could result from credential phishing, password reuse, token theft, malware, session hijacking, social engineering, or another attack path. Suspicious endpoint activity could originate from malicious software, an exploited vulnerability, a compromised administrative tool, or legitimate behavior that resembles an attack.
AI can support root-cause analysis by organizing available evidence and suggesting investigative paths based on what has already been observed.
The analyst might need to determine:
- Were credentials stolen or otherwise compromised?
- Was MFA bypassed or abused?
- Was malware executed?
- Was an exposed or vulnerable service exploited?
- Did the attacker establish persistence?
- Was privilege escalation attempted or achieved?
- Which resources were accessed after the initial compromise?
- Was lateral movement observed?
- Was sensitive information accessed or transferred?
- What was the earliest evidence of malicious activity?
AI can help identify evidence relevant to each hypothesis and point the analyst toward additional data that may confirm or contradict it. This can make investigations more systematic, particularly when the incident spans multiple security domains.
But suggested explanations should remain hypotheses until they are supported by evidence. Generative systems can infer relationships that appear plausible without being correct. In security operations, a convincing explanation is not the same thing as a proven one.
The Tier 2 analyst's role therefore becomes increasingly important as the AI provides more assistance. Someone still needs to distinguish what the evidence shows from what the system believes the evidence might mean.
v. Accelerating Containment and Response Decisions
Once an investigation establishes that malicious activity is occurring, the priority changes from understanding the incident to limiting its impact. Speed becomes critical, particularly when an attacker still has access to an account, endpoint, cloud resource, or network.
AI-assisted workflows can help analysts determine which containment actions are appropriate based on the entities and activity involved. Depending on the organization's security architecture and policies, those actions might include:
- disabling or restricting an account
- isolating an endpoint
- revoking active sessions or tokens
- forcing a credential reset
- blocking malicious infrastructure
- quarantining a file
- creating remediation tickets
- notifying relevant security or IT teams
- preserving evidence for further investigation
- escalating the incident to incident response or Tier 3
The important distinction is between recommending an action and executing an action.
An AI system suggesting that an endpoint should be isolated carries relatively limited operational consequence if an analyst reviews the evidence before approving the recommendation. Automatically isolating a production server based on an incorrect interpretation can have a much larger business impact.
Organizations therefore need different automation thresholds for different actions. Low-risk, reversible actions may be appropriate for automatic execution under clearly defined conditions. Higher-impact actions may require analyst approval. Actions involving critical infrastructure, privileged accounts, business-sensitive systems, or uncertain evidence may require additional authorization.
The objective is not maximum automation.
It is appropriate automation based on confidence, consequence, and risk.
vi. Improving the Tier 1 to Tier 2 Handoff
AI can also improve Tier 2 investigations before the Tier 2 analyst even becomes involved. If Tier 1 enrichment, triage, and documentation are partially automated, escalated incidents can arrive with considerably more context than a traditional ticket containing a short analyst note.
A strong AI-assisted handoff might include the original detection, affected entities, enrichment results, related alerts, relevant historical activity, Tier 1 investigative steps, supporting evidence, unresolved questions, and the reason escalation criteria were met.
That matters because poor handoffs create duplicated work. If Tier 2 cannot understand what Tier 1 investigated or why an incident was escalated, the analyst often has to repeat much of the original triage before moving forward.
AI can help create continuity between the tiers. Tier 1 performs and validates initial triage, the system preserves the investigative context, and Tier 2 continues from that point rather than effectively restarting the investigation.
The result is not merely faster analysts. It is a more continuous investigative process.
The New Tier 2 SOC Analyst
AI can shift Tier 2 away from spending disproportionate amounts of time searching for, assembling, and organizing evidence. The analyst can instead devote more attention to the parts of incident response that require deeper reasoning: determining scope, testing competing hypotheses, understanding attacker behavior, assessing business impact, choosing containment actions, and deciding when more specialized expertise is required.
That does not make Tier 2 less technical. In many ways, it raises the standard.
If AI can retrieve logs, correlate events, construct timelines, and suggest investigative paths, the analyst's value increasingly depends on whether they can determine when those conclusions are incomplete, misleading, or wrong. Tier 2 analysts need to understand the underlying systems well enough to challenge the AI rather than simply accept its output.
The role therefore moves from finding information to interrogating evidence.
Tier 1 asks whether an alert deserves deeper investigation. AI can help answer that question faster.
Tier 2 has a harder responsibility:
What actually happened, how far did it go, what should we do now, and how confident are we in that conclusion?
AI can accelerate the journey toward those answers.
The analyst remains responsible for making sure they are the right ones.
How AI Transforms Tier 3 SOC Analysts
Tier 3 represents the highest level of technical expertise in many SOC structures. These analysts are typically brought into investigations where the available evidence is incomplete, attacker behavior is unusual, the potential impact is significant, or established playbooks no longer provide a sufficient answer. They may be dealing with sophisticated intrusions, previously unseen techniques, difficult forensic questions, or weaknesses in the organization's existing detection capabilities.
That makes Tier 3 considerably harder to automate than routine alert triage. The work often depends on forming hypotheses, recognizing subtle patterns, understanding how technologies behave in a specific environment, and making judgments when there is no predetermined investigative path to follow.
But those same characteristics also make AI potentially valuable as an analytical force multiplier.
Tier 3 analysts frequently work with enormous amounts of information. The challenge is rarely a lack of security data. It is finding the relevant evidence within that data, connecting it across systems, testing possible explanations, and determining what the organization has not yet discovered. AI can help analysts search and organize that information more efficiently, allowing scarce senior expertise to be applied to the reasoning rather than the mechanics surrounding it.
i. Accelerating Threat Hunting
Threat hunting differs from conventional alert investigation because the analyst may begin without a specific detection telling them where to look. Instead, the hunter starts with a hypothesis about how an attacker could be operating and searches available telemetry for evidence that supports or contradicts it.
A hypothesis might be:
Could an attacker be using legitimate administrative tools to move laterally through the environment without triggering existing detections?
Answering that question can require searching endpoint telemetry, authentication records, process execution, network connections, administrative activity, and historical security data across potentially thousands of systems. The analyst needs to translate the hypothesis into searches, examine the results, refine the question, investigate anomalies, and repeatedly narrow the scope.
AI can accelerate this cycle. It can help translate natural-language investigative questions into queries, suggest relevant data sources, identify entities or behaviors worth examining, summarize large result sets, and help analysts refine subsequent searches based on what the evidence reveals.
The value is not that AI independently decides whether an attacker exists. It is that a threat hunter can potentially test more hypotheses across more data in the same amount of time.
That changes the economics of proactive security investigation. Questions that might previously have required substantial manual query development and data review become easier to explore. Analysts can move more quickly from an initial hypothesis to evidence and then use their expertise to determine whether the pattern represents malicious behavior, legitimate administration, or something requiring further investigation.
ii. Supporting Advanced Incident Investigation
Major incidents can produce an overwhelming amount of evidence. A sophisticated intrusion may involve multiple identities, endpoints, applications, cloud resources, network connections, processes, files, domains, authentication events, and security alerts. The investigation may also need to incorporate threat intelligence, previous incidents, vulnerability information, forensic artifacts, and historical telemetry.
At Tier 3, the challenge is often not finding one suspicious event. It is reconstructing the broader attack.
AI can help organize this evidence and identify relationships that deserve further investigation. A Tier 3 analyst could use an AI-assisted system to explore questions such as:
- Where is the earliest evidence of compromise?
- Which identities and systems are connected to the activity?
- What did the attacker access after the initial compromise?
- Is there evidence of persistence or lateral movement?
- Where else does the same behavior appear?
- Have we investigated similar activity before?
- Which detections fired, and which stages of the attack went undetected?
The system can help retrieve relevant evidence, construct timelines, identify shared entities, compare activity across systems, and summarize findings. As the investigation develops, the analyst can continue asking increasingly specific questions rather than manually constructing every search from the beginning.
This is particularly valuable when the scope of an incident changes. An investigation that begins with one compromised endpoint may uncover a stolen identity, additional affected systems, cloud activity, persistence mechanisms, and earlier evidence that shifts the suspected start of the attack. AI can help analysts repeatedly expand and reorganize the investigative picture as new evidence appears.
The Tier 3 analyst, however, remains responsible for determining what that evidence means. Sophisticated attacks frequently contain ambiguity, deliberate deception, incomplete telemetry, and legitimate activity occurring alongside malicious behavior. AI may surface correlations or propose explanations, but those relationships still need to be tested.
AI becomes a force multiplier for the investigator, not the investigator of record.
iii. Understanding Attacker Behavior More Quickly
Advanced investigations often require analysts to move beyond individual indicators and understand attacker behavior. An IP address can change. A domain can disappear. Malware can be modified. What may be more useful is understanding the techniques an adversary used to gain access, establish persistence, elevate privileges, evade defenses, move laterally, and achieve their objective.
AI can help organize observed activity into behavioral patterns and assist analysts in comparing those behaviors with known attack techniques or previous incidents. It may help identify that several seemingly unrelated events could represent different stages of the same intrusion or highlight a technique that was not detected directly but is suggested by the surrounding evidence.
This can help Tier 3 analysts move from indicator-level investigation toward behavior-level investigation more quickly.
The distinction matters because sophisticated adversaries deliberately change infrastructure and tooling. Security teams that understand the underlying behavior can potentially develop more resilient detections than teams relying primarily on individual indicators of compromise.
Human expertise remains critical here. Mapping activity to an attacker technique does not prove attribution, and similar behaviors can appear across different threat actors, legitimate administrative activity, penetration testing, and internal operations. AI can help structure the analysis, but experienced analysts still need to separate plausible interpretations from conclusions supported by evidence.
iv. Improving Detection Engineering
One of the most important outcomes of a serious investigation is not simply resolving the incident. It is ensuring that the organization becomes better at detecting similar activity in the future.
Tier 3 analysts often contribute to detection engineering by examining incidents, identifying behaviors that existing controls missed, developing new detection logic, refining noisy rules, and testing whether security telemetry provides sufficient coverage.
AI can accelerate several parts of this process. It can assist with:
- analyzing missed or partially detected incidents
- identifying common characteristics across related events
- suggesting potential detection logic
- translating detection concepts between supported query languages
- reviewing existing detection rules
- identifying possible coverage gaps
- explaining complex detection logic
- generating test cases
- documenting the purpose and assumptions behind a detection
Consider an investigation where an attacker successfully used a legitimate administrative tool for lateral movement. The SOC may have collected the relevant telemetry but lacked a detection that distinguished suspicious use from normal administration. AI could help the analyst examine historical instances of the behavior, identify potentially useful contextual signals, and draft possible detection logic.
The experienced analyst still needs to determine whether that detection is operationally useful.
A rule that detects every use of a common administrative tool may technically identify the attack behavior, but it could also generate so many false positives that the detection becomes worthless. Effective detection engineering depends on understanding baselines, exceptions, business processes, telemetry quality, attacker tradecraft, and the organization's tolerance for false positives.
AI can accelerate the process of creating and refining detections. It cannot remove the need to understand what meaningful malicious behavior looks like inside a particular environment.
v. Finding Detection and Visibility Gaps
AI can also help Tier 3 analysts ask a broader question after an incident:
Why did we not see this earlier?
An investigation may reveal that the organization had the necessary telemetry but lacked the right detection. In another case, the relevant logs may never have been collected. A detection may have fired but received a low severity rating. Multiple alerts may have existed but were never correlated. An earlier event may have been closed as benign because the context needed to understand it was not available at the time.
By comparing the reconstructed attack with the organization's existing controls, AI can help analysts identify points where detection or investigation broke down.
This creates a useful feedback loop:
Attack activity → Investigation → Detection gaps → Improved detections → Better future visibility
Over time, AI-assisted analysis can potentially help SOC teams learn more systematically from incidents rather than treating each investigation as an isolated event.
That is an important part of Tier 3's value. Senior analysts do not only resolve difficult incidents. They help make the entire security operation better at recognizing the next one.
vi. Making Institutional Security Knowledge More Accessible
Senior security analysts accumulate something that is difficult to represent in a dashboard: organizational memory.
They know which systems behave strangely during routine operations. They recognize recurring false positives. They remember incidents that occurred two years ago and the obscure configuration issue that caused them. They understand which service accounts behave differently from normal users, which applications generate unusual traffic, and which investigative paths have previously produced useful results.
Much of that knowledge exists in people's memories, old incident tickets, investigation notes, internal documentation, playbooks, detection repositories, and scattered security records. When a senior analyst leaves the organization or is unavailable during an incident, some of that context can become difficult to recover.
AI systems connected to approved internal security knowledge can make that information easier for the wider SOC to access.
A Tier 1 analyst investigating unusual behavior might ask:
"Have we investigated activity like this before?"
Instead of manually searching through years of tickets, the system could surface similar historical incidents, summarize what was discovered, identify whether the same systems or accounts were involved, and explain how those cases were ultimately resolved.
A Tier 2 analyst could ask whether a particular behavior has previously been associated with a known internal application. A Tier 3 analyst could retrieve earlier investigations involving similar techniques and compare them with the current incident.
This does not replace senior expertise. It helps capture and distribute some of the knowledge that senior analysts have already created.
There are important controls required around this capability. Security incident records can contain highly sensitive information, so access should reflect existing permissions. Source material needs to be trustworthy and current. Analysts should also be able to see where an AI-generated answer came from rather than treating an unsupported summary as organizational fact.
Used carefully, however, institutional knowledge becomes less dependent on knowing which person to ask.
The New Tier 3 SOC Analyst
AI is unlikely to make the Tier 3 analyst less important. If anything, it can make scarce senior expertise more valuable by reducing the amount of that expertise consumed by mechanical research.
Threat hunters can test more hypotheses. Incident responders can navigate larger volumes of evidence. Detection engineers can develop and refine logic faster. Senior analysts can investigate relationships across broader datasets while making their accumulated knowledge more accessible to the rest of the SOC.
The work that remains is precisely the work that is hardest to automate: recognizing novel behavior, challenging assumptions, understanding incomplete evidence, designing investigative strategies, reasoning about sophisticated adversaries, and deciding how the organization's defenses should change as a result.
This creates a different role for AI at Tier 3 than at Tier 1.
At Tier 1, AI can automate and accelerate much of the repetitive work surrounding triage.
At Tier 2, AI can help assemble and correlate the evidence required to understand an incident.
At Tier 3, AI increasingly becomes a way to extend the analytical reach of the organization's most experienced security professionals.
The Tier 3 analyst therefore moves further toward the work where senior expertise creates the greatest value:
threat hunting, adversary analysis, complex investigation, detection strategy, and continuously improving the organization's ability to detect and respond to threats.
Before and After AI: How SOC Roles Change
Across all three SOC tiers, the pattern is similar. AI takes on more of the repetitive work involved in collecting, organizing, enriching, and retrieving security information, while analysts move toward work requiring interpretation, judgment, and deeper security expertise.
The important change is not simply that every tier becomes faster. The work begins moving up the value chain.
Tier 1 spends less time performing repetitive lookups and more time determining whether the evidence supports escalation. Tier 2 spends less time assembling timelines and more time understanding attack scope and deciding how to contain it. Tier 3 gains tools for exploring larger datasets, testing more hypotheses, improving detections, and investigating sophisticated threats.
In other words, AI can reduce the amount of analyst time consumed by the mechanics of an investigation without eliminating the need for the expertise required to interpret it.
That raises a larger question: if AI can perform more of the work that traditionally happens between SOC tiers, does the SOC still need such rigid tiers at all?
Does AI Eliminate the Traditional SOC Tier Model?
AI may not eliminate the tier model, but it could make it considerably less rigid.
The traditional Tier 1 → Tier 2 → Tier 3 structure exists partly because investigations require progressively deeper expertise. But it also exists because security teams need a practical way to filter large volumes of work. Tier 1 absorbs routine alerts so Tier 2 and Tier 3 do not spend their days investigating every low-risk security event generated by the environment.
That filtering function becomes particularly important when senior security expertise is scarce. If every alert were sent directly to an experienced incident responder, the economics of the SOC would quickly become unsustainable. The tier structure creates a funnel through which events are progressively enriched, investigated, and escalated until the remaining problem justifies deeper expertise.
AI introduces another layer into that funnel.
If an AI-assisted system can enrich an alert, correlate related events, retrieve historical context, execute approved investigative steps, assess the results against defined criteria, and prepare an incident summary before a person becomes involved, the SOC does not necessarily need every investigation to travel sequentially through three human layers.
Instead of:
AI-assisted investigation can potentially reduce those delays by performing work in parallel and moving context with the incident. Enrichment does not necessarily have to wait for an analyst. Searches across connected data sources can begin immediately. Related events can be correlated while the alert is still being processed. An investigative summary can be continuously updated as additional evidence appears.
The future SOC may therefore be less defined by a rigid sequence of analyst tiers and more by dynamic allocation of human expertise.
The question changes from:
"Which tier should investigate this next?"
to:
"What level of human expertise does this incident require right now?"
That is a more significant change than simply making the existing SOC faster.
Will AI Replace SOC Analysts?
AI is likely to automate parts of SOC work. That is not the same as eliminating SOC analysts.
Security operations is unusually resistant to simplistic automation because the environment itself is adversarial. Attackers change techniques when defenses improve. Legitimate administrative behavior can resemble malicious activity. Malicious activity can be deliberately designed to look legitimate. Business systems change, employees behave unpredictably, new applications appear, security controls generate incomplete evidence, and investigations frequently begin without all the information required to reach a confident conclusion.
That creates a fundamentally different problem from automating a predictable administrative workflow.
A security system might determine that an authentication event is statistically unusual, but an analyst may need business context to understand why. An AI system might correlate suspicious activity across an endpoint and identity, but a human may recognize that the behavior resulted from an approved infrastructure change. Conversely, activity that appears individually benign may become concerning because an experienced analyst recognizes a technique, sequence, or organizational context the automated system failed to appreciate.
The consequences of incorrect decisions also matter. Automatically closing a genuine attack as a false positive can allow an adversary to continue operating. Automatically disabling a critical account or isolating an important production system can disrupt the business. The more consequential the action, the stronger the case for appropriate human oversight.
This means the more likely outcome is not the disappearance of SOC analysts but a change in what organizations need analysts to do.
Tier 1 analysts who previously spent large portions of their shifts performing manual enrichment may increasingly supervise and validate investigations that have already been partially assembled. Tier 2 analysts may investigate more incidents because less of their time is consumed by searching across tools and reconstructing basic timelines. Tier 3 analysts may use AI to examine larger datasets, test more threat-hunting hypotheses, develop detections faster, and concentrate on attacks that do not fit established patterns.
SOC managers will also face a new set of responsibilities. They will need to decide which investigative steps can be automated, which response actions require approval, what confidence thresholds are acceptable, how AI-generated conclusions are validated, which systems an AI agent is permitted to access, and how automated decisions are logged and reviewed.
In that sense, AI introduces another security operations problem: the automation itself has to be governed.
A SOC cannot safely assume that an AI-generated investigation is correct simply because it is detailed or persuasive. Analysts need access to the underlying evidence. Automated actions need defined permissions and boundaries. High-impact decisions need appropriate approval mechanisms. AI systems need monitoring for incorrect conclusions, missing context, excessive permissions, and workflows that behave differently from what the security team intended.
There is also a less obvious workforce implication. If AI removes much of the repetitive Tier 1 work traditionally used to train junior analysts, security organizations will need new ways to develop investigative expertise. Entry-level analysts still need to understand logs, endpoint behavior, identity, networks, attacker techniques, and investigative methodology if they are expected to validate AI-generated conclusions. You cannot effectively supervise an investigation you do not understand.
The analyst role therefore does not become less important simply because AI performs more tasks. The value of the human role moves toward the areas where automation is weakest and consequences are greatest:
judgment, verification, investigation, exception handling, adversary reasoning, containment decisions, detection strategy, and security accountability.
The long-term question may therefore be less about whether AI replaces SOC analysts and more about how many routine investigative steps need to remain human once AI can perform them reliably.
That distinction matters.
- AI can automate tasks.
- It can compress investigations.
- It can change escalation paths.
- It may even change the traditional Tier 1, Tier 2, and Tier 3 structure.
But security operations ultimately requires someone to determine whether the evidence can be trusted, whether the conclusion makes sense, and whether the proposed action is appropriate for the organization.
AI can increasingly help reach that decision.
Accountability for the decision is much harder to automate.
AI SOC Automation Needs Guardrails
There is an important distinction between an AI system being capable of recommending an action and an organization deciding that the system should be authorized to execute it.
Consider the difference between:
"The evidence suggests this endpoint may be compromised."
and:
"Immediately isolate this production server from the network."
The first is an analytical conclusion that an analyst can investigate. The second changes the environment and may have immediate operational consequences. If the conclusion is wrong, isolating a critical system could interrupt services, affect customers, disrupt business processes, or interfere with an investigation.
As AI moves from investigation assistance to autonomous action, the level of required control should therefore increase.
This is particularly important in security operations because response actions have very different levels of consequence. Automatically enriching an IP address with threat intelligence is fundamentally different from disabling a privileged account. Generating an incident summary is different from revoking authentication sessions. Searching historical logs is different from blocking network traffic or isolating an endpoint.
Organizations should define clear boundaries around what AI is permitted to do. Those boundaries should address:
- what information and security systems AI can access
- which investigative steps it can perform automatically
- which conclusions it can recommend but not act upon
- which low-risk and reversible actions can be automated
- which response actions require analyst approval
- which systems, identities, or assets require additional authorization
- when uncertainty must trigger human escalation
- how AI-generated findings and actions are logged
- how analysts can inspect the evidence behind a recommendation
- how automated actions can be reviewed, reversed, or interrupted
A useful approach is to think about automation in terms of confidence, consequence, and reversibility.
An action may be a stronger candidate for automation when the evidence is highly reliable, the potential business impact is limited, and the action can easily be reversed. As uncertainty or operational consequence increases, human approval should become more important. Actions involving critical infrastructure, privileged identities, sensitive data, or significant business disruption should generally face a much higher automation threshold than routine investigative tasks.
AI systems also need constrained permissions. An investigation agent does not necessarily need unrestricted authority across every security platform simply because those integrations are technically possible. Access should reflect the tasks the system is expected to perform, following the same principle of least privilege organizations apply elsewhere in their security architecture.
Auditability matters for the same reason. Analysts should be able to determine what information an AI system accessed, which evidence contributed to a conclusion, what recommendation was produced, which automated action occurred, and whether a human approved it. When something goes wrong, the SOC needs to reconstruct the decision just as it would reconstruct the security incident itself.
There should also be an explicit answer to a simple question:
What does the AI do when it does not know?
A well-designed security workflow should be able to escalate uncertainty rather than conceal it behind a confident-sounding conclusion. Missing telemetry, conflicting evidence, unfamiliar behavior, inaccessible systems, and low-confidence interpretations should all be legitimate reasons to involve an analyst.
The objective is therefore not autonomy at any cost.
It is to increase the speed and capacity of security operations while preserving control, evidence, accountability, and appropriate human judgment.
Where SOC Teams Should Start With AI
The strongest starting point for AI in a SOC is usually not the most sophisticated autonomous security use case. It is the repetitive investigative work already consuming significant analyst capacity.
Start by examining the current SOC workflow in detail. Where does analyst time actually go? How much of an investigation involves searching other systems for context? How often are the same enrichment steps repeated? How much time is spent writing summaries and updating tickets? How frequently does Tier 2 repeat work because an escalation arrived without enough information? How long do incidents wait between stages of investigation?
These questions expose where the existing operating model is creating friction.
A SOC may discover, for example, that analysts spend several minutes enriching almost every identity alert with the same information. Another team may find that Tier 2 analysts routinely repeat Tier 1 searches because escalation notes are incomplete. A third may discover that senior analysts spend substantial time translating investigative findings into incident documentation rather than conducting deeper analysis.
Those are strong AI opportunities because there is already a measurable problem to solve.
A useful implementation sequence is:
Alert enrichment → Incident summarization → Investigation assistance → Recommended actions → Controlled automation
The first stages are relatively low risk because AI is primarily gathering and organizing information. Analysts still make the security decisions. This gives the SOC an opportunity to evaluate whether the system retrieves the correct evidence, produces accurate summaries, handles exceptions appropriately, and integrates effectively with existing investigative processes.
Once those capabilities become reliable, AI can move deeper into investigation. It might construct timelines, identify related entities, suggest investigative queries, compare activity with historical incidents, or recommend additional steps based on an approved playbook. Again, the analyst can validate the output before decisions are made.
Recommended response actions represent another step in maturity. The AI might determine that available evidence supports isolating an endpoint, revoking a session, or escalating an incident, while leaving execution to the analyst. This allows the organization to evaluate the quality of AI recommendations before granting the system authority to act on them.
Only after the SOC has established sufficient confidence should it consider autonomous execution of appropriate response actions. Even then, automation does not need to be all or nothing. Some actions may be automated while others continue to require approval.
This creates a progressive model:
Assist → Recommend → Approve → Automate
The advantage of this approach is that trust is earned through observed performance rather than assumed because the technology appears capable.
Before introducing AI, the SOC should also establish a baseline for the workflow being changed. If the objective is faster alert triage, measure current triage times. If analysts are spending too much time gathering context, estimate how much investigative time is devoted to enrichment. If escalation quality is the problem, examine how often Tier 2 has to repeat Tier 1 work.
Without a baseline, it becomes difficult to determine whether AI has actually improved security operations or simply added another layer of technology.
The best first AI project is therefore usually a workflow that is high-volume, repetitive, measurable, sufficiently structured, and low enough in consequence to introduce automation safely.
Solve one operational problem. Measure the result. Improve the workflow. Then expand.
Measuring the Value of AI in the SOC
AI adoption in security operations should ultimately be judged by operational and security outcomes, not by how many AI tools, copilots, or agents the organization has deployed. The fundamental question is whether the SOC can investigate threats faster, more consistently, and with better use of human expertise without introducing unacceptable new risks.
Several existing SOC metrics can help measure that change:
Speed alone should not define success.
A system that reduces mean time to triage by automatically closing more alerts may appear highly effective until the organization discovers that genuine threats are being incorrectly dismissed. Similarly, an AI system that increases alerts handled per analyst is not necessarily improving security if escalation quality deteriorates or analysts spend additional time correcting inaccurate AI-generated investigations.
This is why quality metrics should sit alongside productivity metrics.
Organizations can examine how frequently analysts accept, modify, or reject AI-generated conclusions. They can measure whether AI-assisted escalations contain more complete evidence. They can review the accuracy of incident summaries, compare recommended response actions with analyst decisions, and track how often automated workflows require correction or rollback.
Another useful measure is where analyst time moves after automation.
If Tier 1 analysts spend fewer hours on enrichment, are they spending more time investigating meaningful alerts? If Tier 2 saves time constructing timelines, does that translate into faster scoping and containment? If Tier 3 spends less time searching historical incidents, are analysts conducting more threat hunts or improving detection coverage?
This matters because the real objective is not merely to reduce the number of minutes associated with individual tasks. It is to redirect scarce security expertise toward higher-value security work. A mature measurement framework should therefore consider three dimensions:
Speed: Are investigations and responses happening faster?
Quality: Are decisions, escalations, and investigations becoming more accurate and useful?
Capacity: Can the existing team effectively investigate more security activity without an equivalent increase in repetitive human workload?
Those measurements reveal whether AI is actually changing SOC performance. Because ultimately, the strongest business case for AI in security operations is not that the SOC has adopted AI. It is that the SOC can detect, investigate, and respond to threats more effectively because of it.
The Future SOC Analyst Will Work With AI
AI is not simply another security product to add to an already crowded SOC stack. Its more significant impact is that it can change how security work is distributed between people and machines.
- Tier 1 analysts can move away from spending large portions of their shifts manually enriching and processing alerts toward validating findings, assessing risk, handling exceptions, and making better escalation decisions.
- Tier 2 analysts can spend less time reconstructing incidents from fragmented evidence and more time determining scope, understanding attacker behavior, and coordinating containment.
- Tier 3 analysts can use AI to search larger datasets, test more hypotheses, accelerate threat hunting, and improve detection engineering.
Across all three tiers, the direction is the same. AI takes on more of the mechanical work required to prepare an investigation, while analysts concentrate on the decisions that require security expertise.
That does not point inevitably toward a SOC without analysts. It points toward a SOC where human attention is used more deliberately.
This distinction becomes increasingly important as security environments become more complex. Organizations continue to generate more telemetry, deploy more security tools, operate across more cloud services, and defend identities, endpoints, applications, and infrastructure simultaneously. Simply adding analysts to keep pace with every increase in security data is difficult to sustain.
The future SOC analyst is therefore unlikely to compete with AI by trying to process alerts faster than a machine. The analyst's value will increasingly come from what machines still struggle to provide reliably: understanding ambiguity, challenging conclusions, recognizing novel behavior, weighing operational consequences, applying business context, and taking responsibility for consequential security decisions.
AI can provide investigative capacity.
Human expertise determines how that capacity should be used.
Build AI Into Security Operations With Shift AI
At Shift AI, we approach AI automation from the workflow outward.
Rather than starting with the question, "Where can we add AI?", we start by examining where security teams are losing time today. That might be repetitive alert enrichment, manual investigation steps, fragmented information across security systems, incident documentation, internal knowledge retrieval, or escalation processes that repeatedly require analysts to gather the same context.
For cybersecurity teams, MSPs, and MSSPs, potential workflows can include AI-assisted triage, alert enrichment, investigation support, incident summarization, internal security knowledge retrieval, escalation workflows, and other repetitive operational processes surrounding the SOC.
The objective is not to remove security expertise from the process.
It is to make that expertise more scalable.
That means identifying where AI can safely gather information, perform repetitive investigative steps, prepare evidence, or recommend actions, while keeping appropriate huma







.png)

%20(600%20x%20600%20px)%20(7)%201.png)

%20(600%20x%20600%20px)%20(3)%201.png)
%20(600%20x%20600%20px)%20(2)%201.png)
%20(600%20x%20600%20px)%20(4)%201.png)
%20(600%20x%20600%20px)%20(6)%201.png)
.png)
.png)
.png)


