AI for Cybersecurity: AI Agents That Triage Alerts, Chase Vulnerabilities and Collect Audit Evidence (2026)
AI for cybersecurity now covers two jobs: detection inside your security tools, and the operational work around them that eats a security team's week. This guide explains both, shows the workflows SPIRITT runs for security teams, how security AI tools are priced in November 2026, and the controls to demand before any AI agent touches your environment.
AI for cybersecurity is software that detects threats faster and takes the repetitive security work off your team: triaging alerts, enriching indicators, chasing vulnerability fixes, running access reviews, answering security questionnaires and collecting audit evidence. SPIRITT is an AI agent platform security teams use for that operational work: it connects to your ticketing, cloud, identity and communication tools, does each step, and asks a person before any change, with every action logged. Detection engines inside your EDR, SIEM and email gateway already use machine learning to spot attacks. What most teams still do by hand is everything around those alerts. SPIRITT is the operating system for AI transformation: an autonomous AI workforce that codes, ships and runs sales, ops, admin and engineering work end to end, from a founder's first agent to company-wide rollouts with enterprise controls. For a security team that means an agent that writes the enrichment script, builds the dashboard, files the tickets and follows up until each item is closed, working on its own isolated cloud computer around the clock.
How is AI used in cybersecurity?
AI is used in cybersecurity in two layers, and buyers mix them up all the time.
- Detection and response AI inside security products. Machine learning models in endpoint, network, identity and email tools flag unusual behavior, score alerts and sometimes isolate a device. Vendors like Microsoft, CrowdStrike, Palo Alto Networks and Google build this into their platforms, and assistants such as Microsoft Security Copilot and CrowdStrike Charlotte AI summarize incidents and answer questions about the data in those platforms.
- AI agents for security operations. These agents work across tools rather than inside one. They read the alert in the SIEM, look up the user in the identity provider, check the asset in the cloud console, open the ticket in Jira or ServiceNow, message the owner in Slack and follow up. This is where SPIRITT works.
The first layer finds the problem. The second layer gets it handled. Most security teams are short on people for the second layer, which is why the backlog of open findings grows even when detection is excellent.
What security teams use AI agents for
| Security task | By hand | With an AI agent | Who approves |
|---|---|---|---|
| Alert triage | Open each alert, pivot through four consoles | Enriches the alert with user, asset, recent logins and threat intel, drafts a verdict and next step | Analyst confirms the verdict |
| Vulnerability follow-up | Export scanner results, email owners, chase in spreadsheets | Groups findings by owner, files tickets with due dates by severity, reminds owners, verifies fixes on the next scan | Security lead sets the SLA |
| Phishing reports | Read every reported email | Checks headers, links and attachments in a sandbox service, replies to the reporter, escalates real ones | Analyst on escalations |
| Access reviews | Pull user lists, send to managers, collect answers | Builds the review list from the identity provider and HR system, sends to managers, records answers, drafts removals | Manager and IT admin |
| Security questionnaires | Copy answers from last year's file | Drafts answers from your policies and past questionnaires, flags anything new | Security or compliance owner |
| Audit evidence | Screenshot settings before every audit | Collects evidence on a schedule (MFA coverage, backups, logging, offboarding) into a dated folder | Compliance owner |
| Security reporting | Build the monthly deck by hand | Pulls metrics from each tool and drafts the report | CISO |
Six workflows SPIRITT runs for a security team
1. Alert enrichment before an analyst opens it
Ask once: "When a high-severity alert lands, enrich it before anyone looks." SPIRITT reads the alert, looks up the user in Okta or Microsoft Entra ID, pulls recent sign-ins and device details, checks the IP or hash against the threat intelligence sources you use, and posts a short summary with a suggested verdict in the incident channel. Where your tools have an API it uses the API; where they do not, it uses the console on its own cloud computer, like a person would. The analyst starts from a filled-in case, and the containment step waits for a human click.
2. Vulnerability findings that actually get closed
Scanner exports are long and owners are many. SPIRITT takes the findings from your scanner, maps each host or repository to an owner, groups the work, files one ticket per owner in Jira, ServiceNow or Linear with due dates by severity, and follows up in Slack or Teams before each date. On the next scan it checks which findings are gone and closes those tickets with a note. You get a weekly list of what is overdue and who owns it.
3. Phishing inbox triage
Employees report suspicious email to a shared mailbox. SPIRITT reads each report, checks sender records, links and attachments with the analysis services you approve, answers the reporter ("This one is safe, thanks for checking" or "Good catch, we are on it"), and escalates real phishing with the evidence attached. Blocking a sender or purging messages stays behind approval.
4. Quarterly access reviews
SPIRITT builds the review from the identity provider and your HR system, sends each manager the list of people and permissions to confirm, collects the answers, reminds the late ones and drafts the removal tickets. It flags accounts with no login in 90 days and contractors past their end date. IT approves each removal.
5. Security questionnaires and trust requests
Sales teams wait on security questionnaires. SPIRITT drafts answers from your policies, your last SOC 2 report summary and past questionnaires, marks anything it could not answer from a source, and sends the draft to the security owner. Answers that change get written back to a single answer library, so the next questionnaire starts further along.
6. Continuous audit evidence
Instead of a scramble before the audit, SPIRITT collects evidence on a schedule: MFA coverage, admin role lists, backup job results, logging settings, offboarding records. Each item lands in a dated folder with a short note on what was checked. When the auditor asks, the evidence is already there.
Because SPIRITT also writes and runs code, it can build what your process is missing: a small enrichment service, a parser for a legacy log format, a dashboard that joins scanner and ticket data. It composes models within each task for the optimal mix of capability and price and connects to 1,500+ apps, so it fits the stack you already have.
A worked example: Monday morning in a five-person security team
A made-up mid-size company runs Microsoft Defender, Okta, a cloud vulnerability scanner and Jira. Over the weekend:
- 14 high-severity alerts came in. SPIRITT enriched each one; 11 matched known admin activity and were drafted as benign with the evidence, 3 were escalated with user, device and sign-in history attached. The on-call analyst confirmed all 14 verdicts in 25 minutes.
- The weekly scan found 212 findings. SPIRITT matched 198 to existing tickets, filed 9 new tickets for 6 owners, and closed 31 tickets whose findings were gone.
- Nine phishing reports arrived. Seven were marketing email; reporters got a thank-you reply. Two were credential phishing; the agent attached the evidence and asked for approval to purge the messages from all mailboxes.
The team spends Monday on the three real alerts and the two phishing campaigns, not on sorting.
Security AI tools compared on price (November 2026)
| Tool | What it does | Works across your tools | Price (US, November 2026) |
|---|---|---|---|
| SPIRITT | AI agent platform that runs security operations work: triage, vulnerability follow-up, access reviews, questionnaires, evidence, reports; writes the scripts and dashboards you need | Yes, 1,500+ apps plus custom integrations, with approvals | Free to start; Starter $32, Pro $256, Max $1,024 a month; Enterprise custom |
| Microsoft Security Copilot | AI assistant and agents inside Microsoft Defender, Entra, Intune and Purview | Microsoft security stack plus plugins | Included for eligible Microsoft 365 E5 and E7 customers (400 SCUs a month per 1,000 licenses, up to 10,000); otherwise $4 per provisioned SCU per hour and $6 per overage SCU |
| CrowdStrike Charlotte AI | AI analyst inside the Falcon platform | Falcon data and connected sources | No published price |
| ChatGPT Business | General AI assistant for drafting and analysis | Through connectors and uploads | $20 per user a month billed annually |
The tools fit together. A Microsoft shop keeps Security Copilot for investigation inside Defender and uses SPIRITT for the cross-tool work: the tickets, the follow-ups, the reviews and the evidence.
Controls to demand before an AI agent touches security tools
- Isolation. Every SPIRITT workspace runs on its own isolated cloud computer. SPIRITT is a Google Cloud launch partner for Agent Substrate, which runs agent code in gVisor isolation on GKE (Google Cloud blog, November 5, 2026).
- Keys the agent never sees. Credentials are injected at the network boundary, so the agent uses a connection without reading the secret.
- Read, Write and Destructive rules per integration. Allow, ask or never, with single-use approvals. Reading alerts can be automatic while isolating a host always asks.
- A full audit trail. Every action, by which agent, on which system, is logged.
- SSO and roles. SAML or OIDC single sign-on, SCIM and scoped roles. SOC 2 Type II is listed on the SPIRITT enterprise page.
- Deployment where your data must stay. SPIRITT cloud, your own cloud, your servers or fully air-gapped.
- Cost limits. Budgets per organization, team, project, agent and task, with alerts, hard stops and itemized receipts.
More detail is in SPIRITT security and the most secure AI agent platforms for enterprises.
How to start
- Sign up free and describe one job: "Every Monday, group open scanner findings by owner and file tickets."
- Connect read-only access to one tool first, or start with the made-up test data in the prompt below.
- Review the first run with the agent and adjust the rules.
- Put it on a schedule, then add the next job: phishing triage, access reviews or evidence collection.
IT teams running the helpdesk side can read AI agent for IT support, and compliance owners can read AI for compliance.
FAQ
How is AI used in cybersecurity?
AI is used in cybersecurity to detect threats inside security products (endpoint, network, identity and email tools score unusual behavior) and, increasingly, as AI agents that run the operational work across tools: alert enrichment, vulnerability follow-up, phishing triage, access reviews, security questionnaires and audit evidence.
What are the best AI tools for cybersecurity?
The right mix is usually one detection platform plus one agent layer. Detection and investigation live in tools like Microsoft Defender with Security Copilot or CrowdStrike Falcon with Charlotte AI. SPIRITT is an agent platform for the cross-tool operations work, with approvals and an audit trail on every action.
What is an AI agent for cybersecurity?
An AI agent for cybersecurity takes a goal such as "make sure every critical finding has an owner and a ticket" and does every step: reads the scanner, maps owners, files tickets, follows up and verifies the fix, asking a person before any change to production systems.
Can AI agents respond to incidents on their own?
They can, within the rules you set. A sensible setup lets the agent enrich, summarize and draft the response automatically, while containment actions such as isolating a host, disabling an account or purging email wait for a single-use human approval.
Will AI replace cybersecurity analysts?
AI takes over enrichment, sorting, chasing and first drafts. Judgment on real incidents, threat hunting, architecture and accountability stay with the analysts, who now spend more of their week on the alerts that matter.
How much does AI for cybersecurity cost?
SPIRITT is free to start, then Starter $32, Pro $256 or Max $1,024 a month on credits, with Enterprise priced custom. Microsoft Security Copilot is included for eligible Microsoft 365 E5 and E7 customers, or $4 per provisioned SCU per hour otherwise.
Is it safe to connect an AI agent to security tools?
Yes, with the right controls: isolated execution, keys the agent never sees, per-action allow, ask or never rules, single-use approvals, a full audit trail, SSO and the option to deploy in your own cloud or air-gapped. SPIRITT has all of these.
How we checked
SPIRITT publishes this guide. SPIRITT capabilities, controls, deployment options and prices come from spiritt.ai, the SPIRITT enterprise page and the live SPIRITT plan catalog, checked in November 2026. Microsoft Security Copilot pricing and the Microsoft 365 E5 and E7 inclusion come from microsoft.com/security/pricing and Microsoft Learn; CrowdStrike lists no Charlotte AI price, which independent price reviews from November 2026 confirm; ChatGPT Business pricing comes from openai.com; all were checked on November 21, 2026. We reviewed the top Google results for "AI for cybersecurity" on November 22, 2026 to cover the questions searchers ask. The Monday example is a composite with made-up numbers, not a real company.