AI agents investigate every alert, add context, and prepare cases for response, on a full-stack SecOps platform that captures and correlates all your security data.
Cloud-based SIEM platform providing visibility across all your data sources
Every alert investigated, with a verdict and the evidence behind it
Manage your external footprint, publicly available data, and Dark Web leaks
Log management and threat detection across any of your log sources
Get ultimate assurance that your environment is free from threat actors
Fully featured distrubuted NDR for on-premises and public cloud environments
A SecOps platform purpose-built for industrial and OT environments
Integrated endpoint agent providing deep visibility, response and forensics
Deploy honeypots and honeytokens and stay ahead of threat actors
24x7 security operations from our Australian SOC. AI investigates every alert, and experienced analysts validate findings and contain threats around the clock.
24x7 managed detection and response across endpoint, network, cloud, identity, and log sources
AI investigations and 24x7 MDR on the Microsoft Sentinel you already own
Eyes-on-glass SOC and continuous threat hunting
Expert hunters and AI agents find attacker activity your alerts missed
Quickly recover from cyber attacks and breaches
ThreatDefence is the only SecOps as a Service company providing broad coverage across your entire technology stack with evidence-based security.
We provide security teams with full-stack SecOps infrastructure – you get deep visibility from day one, gain control over your security data, and get access to a comprehensive set of ready-to-use tools, detections, workflows, playbooks and scenarios.
24x7 SOC, Continuous Monitoring, Threat Containment and Incident Response
Visibility and actionable insights for all IT assets
Leverage our Cyber Range to train defenders based on real-world scenarios and simulations
OT and industrial cyber security services for critical infrastructure
Monitoring, evidence and a 24×7 SOC in one platform
Compliance monitoring, evidence and board reporting
Privacy Act reform impacts, obligations, and practical security considerations
Monitoring and compliance support for critical infrastructure environments
A practical guide to securing critical infrastructure with continuous monitoring
A practical buyer’s guide for schools evaluating SIEM and SOC services
A practical guide to selecting the right SIEM and SOC service for local government
Articles on security operations, AI and compliance
Events, awards and company news
Read what our customers say about us
Solution briefs and datasheets
How we keep your data secure
Technical support and knowledge base
Guides, case studies, and advisories in one place
How AI agents investigate alerts and run security operations
Free ebook: a practical guide to adopting an AI SOC platform
A practical guide to incident response and digital forensics
Where SOAR playbooks fit and where AI agents take over
Free ebook: modern security operations with Microsoft Sentinel
How the framework is structured and what changes from the Essential Eight
How an attack unfolds stage by stage and how to detect each one
November 18, 2026
November 16, 2026
November 13, 2026
We provide end-to-end SecOps solutions for Managed Security Service Providers (MSSPs), Managed Service Providers (MSPs), and Value-Added Resellers (VARs).
Find out how you can become a ThreatDefence partner.
Protect your entire technology stack and build your SecOps on your own pace
White-labeled AI SOC for every client tenant, co-managed with our SOC team
Launch your SOC business without any upfront investment
Embed security operations, threat intelligence and exposure data in your product
Sovereign SOC for Australian MSPs and MSSPs, with onshore analysts and client data
Set the rules for closing alerts automatically, with evidence checks and review
How MSPs deliver branded security monitoring with a SOC partner
ThreatDefence was an official partner of the Cyber Resilience Summit NSW, 29 October 2026 at Hilton Sydney.
Our team was back at BSides Canberra, and our cyber quiz was a hit again.
Two days in Perth with our partner Ever Nimble, on security operations and cyber resilience.
Sponsors of the AUSCERT Cyber Security Conference on the Gold Coast.
Sponsors of the Gartner Security & Risk Management Summit in March.
Sponsors of the Cyber Security for Critical Assets Summit in Perth, February 2026.
Expert conversations with speakers and attendees at CyberCon Melbourne 2025.
AI incident response is the use of AI agents and automation to carry out parts of a security incident response. Agents gather evidence, connect events, draft findings, and prepare or perform containment actions, while people make the decisions with business consequences.The clearest way to see what changes is to follow one alert handled both ways. This guide does that, then goes through response tasks one by one and says which to automate and which to keep with people.
Security operations have a problem of arithmetic. Monitoring tools produce more alerts than people can investigate. Each alert, taken seriously, needs the same set of lookups: who is this user, is this address known, what else did this device do, has this happened before. A thorough investigation of one alert can take an analyst half an hour or more, and a mid-sized organization can generate hundreds of alerts a day.
The usual result is triage by neglect. Analysts investigate the alerts that look worst and close the rest on a glance. Attackers benefit, because early signs of an intrusion rarely look like the worst alert in the queue.
Earlier answers went part of the way. Better tuning reduced the number of alerts and also suppressed some that mattered. Fixed automation, where a tool runs a scripted sequence for each alert type, removed repetitive lookups but could not go beyond its script.
AI agents address the part that remained: the investigation itself. An agent reads each result and continues until it has enough to reach a conclusion. That is the work analysts never had time to do for every alert.
The effect on incident response is at the start of the timeline. The stages that cost the most time in a real incident are noticing, understanding, and scoping. Those are the stages agents accelerate.
At 02:10 an identity alert fires. A finance user’s session, opened from the office at 01:58, has appeared on a second network with a different browser. The sequence below compares an on-call analyst working by hand with an AI agent that investigates and holds actions for approval.
The analyst did nothing wrong in the left-hand column and did the same work in the same order. The difference is who did the gathering and when the person’s judgment was applied. Containment came 45 minutes earlier, and the analyst spent their time deciding instead of collecting.
The agent did not decide to wake the finance manager, and it did not decide what to tell the two colleagues. Those stayed with a person. The containment itself follows the order set out in the account takeover guide: block, reset, revoke, remove, confirm.
An agent is given a goal, such as deciding whether a sign-in is malicious, and a set of tools it is allowed to use. Each tool is a defined action: search sign-in logs for a user, look up an address, list processes on a host, fetch the audit trail for a mailbox. The agent then works in a loop.
The design of the investigation side determines whether the output can be trusted.
A well-built agent cites the log entries behind each statement in its case. A conclusion without evidence should be treated as unverified.
Investigation tools retrieve data and change nothing. Actions that alter systems are separate, fewer, and governed by the levels of automation described later.
A cap on steps, time, and cost per investigation stops an agent from wandering.
Knowledge about the environment, such as which accounts are administrators or which behavior has been approved before, is supplied deliberately and can be reviewed.
The underlying language model supplies reasoning. The tools supply facts. An agent is only as good as the data its tools can reach, and that is the first thing to check in any deployment.
Automation in response is set per action. Each action type can sit at a different level, and the level can rise as the record supports it.
Set the level per action type and per part of the organization. Revoking a session for a standard user might be autonomous. Isolating a production server might wait for approval. Disabling the chief executive’s account might stay at co-pilot.
In the example, containment for a finance account was at wait for approval. At autonomous, it would have happened at 02:13 and the analyst would have been told afterward.
Automation in security operations is not new. Orchestration tools, known as security orchestration, automation, and response (SOAR), have run fixed playbooks for years: when this alert fires, run these steps in this order.
A fixed playbook does exactly what was written. It works well for predictable tasks and stops being useful when an incident does not match the script.
An AI agent chooses its next step from what it has found. In the example, the agent checked other users whose sessions appeared on the same network because the first evidence suggested a stolen session. Nobody wrote that branch in advance. This is what lets an agent handle incidents that fall between playbooks.
Fixed automation suits tasks that should never vary, and agents suit investigation, where the right next question depends on the last answer.
Introducing agents changes the work people do more than it changes how many people are needed.
Analysts review cases instead of building them. The job shifts from gathering evidence to judging it. This is more demanding in one way: an analyst has to be able to tell when a well-written case is wrong.
Decisions concentrate. With the routine investigation done, what reaches a person is the set of cases that need a human call. The proportion of an analyst’s day spent on judgment rises.
New responsibilities appear. Someone has to decide which actions may be automated, review the agent’s work, investigate its mistakes, and maintain the context it relies on. In mature teams this becomes a defined role.
Experience matters more. Junior analysts traditionally learned by doing the repetitive work that agents now take. Teams need a deliberate way to build judgment in new staff, such as reviewing agent cases alongside a senior analyst.
Coverage improves at night and on weekends. An agent investigates the 3 a.m. alert as thoroughly as the 3 p.m. one. The person on call receives a finished case instead of a raw alert, which changes what on-call work is.
The measures that show whether it is working are the time from alert to verdict, the time from alert to containment, the share of alerts investigated in full, and how often people override the agent. The number of alerts closed is a poor guide.
Alert investigation is the most visible use. The same approach applies to other parts of security operations that are limited by analyst time.
A hunt starts from a question, such as whether any account has consented to an unfamiliar application in the past 90 days. An agent can run the searches across months of retained data, follow up on what it finds, and write up the result, including when it finds nothing.
When a new threat is disclosed, an agent can check retained data for its known traces across every monitored environment within hours.
Once one indicator is known, an agent searches every user and host for it, then for whatever those results reveal.
An agent can compare a new advisory with the asset inventory and say which systems run the affected version.
As an agent works alerts, it sees which rules fire on benign activity. It can propose changes, such as narrowing a rule or adding an exception, for an engineer to review.
Monthly summaries and service reviews can be drafted from case records that already contain the evidence.
An automated action that is wrong can cause an outage. Controls keep the risk in proportion to the benefit.
Every automated action has a documented way to undo it, and the undo is tested.
The agent states how sure it is, acts only above a set confidence, and passes uncertain cases to a person.
Rules set which systems and accounts can be acted on, and how many in a given period, so one mistake cannot take down a department.
The agent can reach only the systems and data each task needs.
Text inside logs, emails, and alerts is treated as data. An instruction hidden in a phishing email changes nothing about what the agent does.
People sample automated decisions regularly, and wrong actions are treated as failures to learn from.
No organization should give an agent authority on the strength of a demonstration. The evaluation uses the organization’s own alerts and happens before the agent can change anything.
Questions worth asking any supplier of an agent-based service:
An agent cannot investigate a system that sends no telemetry, or search logs that were never kept.
A finance director signing in from a hotel abroad looks like an attack. Context the agent lacks, such as a travel booking, decides it.
An agent may miss activity that does not resemble anything it has seen.
An agent tuned to act quickly will sometimes act on a false alarm. Rollback and scope limits decide how much that costs.
Coordination, communication, and hard decisions still need experienced people. Automation shortens the technical work and frees time for them.
Automation works on top of a tested process. A procedure that people have never run by hand should not be handed to an agent.
ThreatDefence combines an AI SOC platform with a 24x7 security operations center. AI agents investigate every alert by querying connected security tools, correlating endpoint, identity, network, and cloud evidence, and preparing a verdict with supporting records, the queries performed, and recommended response actions.
Operating modes are configured per customer and action type: co-pilot, wait for approval, or autonomous execution for nominated actions. Every action has a documented rollback, low-confidence cases are escalated to a person, and ThreatDefence analysts review every autonomous action.
Confirmed incidents are escalated to the SOC and DFIR team for response.
Experiencing a breach?
Our incident response team answers 24x7, with or without a retainer.
Explore the AI SOC Platform
See how ThreatDefence automates security monitoring, compliance and cyber risk management in one platform.
CountryAustraliaUnited StatesNew ZealandUnited KingdomCanadaSingaporeOtherI'm an MSP, MSSP, or reseller
Δ
AI incident response puts an AI agent at the front of the response process. When an alert fires, the agent pulls the logs, checks history and context, links related events, and writes a case with a verdict and the evidence behind it, usually within minutes. Containment actions are proposed for approval or run automatically within limits the organization has set. People still own scope, recovery, notification, and anything with legal weight.
No. Investigation and low-impact containment, such as revoking a session or isolating a standard laptop, can run without a person. Taking a server offline, cutting a supplier’s access, deciding that recovery is safe, notifying regulators, and speaking for the organization cannot. Teams that adopt agents keep their responders, whose work moves from collecting evidence to checking cases and making those calls.
Start with tasks that change nothing: pulling logs, adding context, linking events into one case, building the timeline, and searching for the same traces across every user and host. Next come actions that are quick to undo, such as session revocation and isolation of a standard endpoint, first with approval and later automatically. Anything that can stop a business service stays with a person.
SOAR, short for security orchestration, automation, and response, runs playbooks written in advance: a named trigger starts a fixed series of steps. An AI agent has no fixed series. It chooses each query from the result of the last, so it can follow an incident that no playbook anticipated. Most deployments use both, with the agent investigating and a playbook carrying out the approved action the same way every time.
Yes, in a specific way: a wrong action interrupts legitimate work. Blocking a finance director who really is traveling, or isolating a server during payroll, costs more than a missed low-severity alert. The risk is managed by limiting which accounts and systems an agent may act on, capping how many actions it can take in a period, requiring approval above a set impact, and testing the rollback for every action.
An agent needs the same sources an analyst would open: identity sign-in and audit logs, endpoint telemetry, mail and file audit records, and network and cloud logs, with enough history to compare today against the past month. It also needs context that logs do not hold, such as who owns each asset, which accounts are privileged, and what has been explained before. A source the agent cannot query is a blind spot in every case.
Every statement in the case should cite the log record behind it, so a reviewer can open the record and check. Before go-live, the agent is replayed against past incidents and run in shadow mode beside analysts, and each disagreement is traced to who was right. In production, the rates at which people override its verdicts and roll back its actions show whether it still deserves its level of authority.
Incident Response Playbooks
Cyber Incident Response Guide
AI SOC Agents