Microsoft rolls out Security Copilot agents to automate SOC work and make phishing triage 6.5x faster

Microsoft is adding AI agents across Defender, Entra, Purview to cut noise and speed triage, included with M365 E5. 6.5x faster phishing triage, safer policy ops, human sign-off.

Categorized in: AI News Operations
Published on: Nov 20, 2025
Microsoft rolls out Security Copilot agents to automate SOC work and make phishing triage 6.5x faster

Microsoft's new AI security agents: a practical playbook for Operations

Microsoft is embedding more than a dozen AI agents across Defender, Entra and Purview to automate high-volume security work. For Operations leaders, this isn't hype - it's a way to reduce alert fatigue, standardize response, and stretch senior analyst time without adding headcount.

The agents live inside Security Copilot and are included with a Microsoft 365 E5 licence at no extra cost. They act like always-on assistants that run triage, enrich alerts, and handle policy changes with traceable steps and human approval where needed.

What's actually new

  • Phishing triage agent: processes alerts faster with measurable gains - Microsoft reports 6.5x faster throughput, 550% efficiency, up to 77% better accuracy, and analysts spending 53% more time on confirmed cases.
  • Natural-language threat hunting: junior analysts can issue complex hunts in plain English, reducing reliance on a few Level 3 specialists.
  • Identity policy ops (Entra): routine policy checks and exceptions handled consistently with audit trails.
  • Data security (Purview): classification and alert enrichment at scale to surface the highest-risk events first.

Why Operations should care

Attackers are using AI to craft better lures and spin up malware faster. Your response needs speed and consistency without burning out the team. These agents aim to cut noise, improve accuracy, and let analysts focus on investigations that matter.

They also help close skills gaps - Microsoft estimates 4.7 million unfilled cyber roles worldwide - by making advanced workflows easier to run and repeat.

Guardrails and control

  • Human-in-the-loop: analysts can review, approve, and override. Every action is logged for audit.
  • Responsible AI: Microsoft says agents are monitored and aligned to its policy framework. See Responsible AI principles.
  • Scoped permissions: run agents with least-privilege and enforce role-based access for sensitive actions.

Build vs. adopt

Beyond the prebuilt options, customers are already creating custom agents for unique workloads - Microsoft notes 370+ have been built since September. Start with standard agents for quick wins, then clone and adapt once your playbooks stabilize.

30-day pilot plan for your SOC

  • Week 1 - Prep: confirm M365 E5 eligibility; pick one workflow with high volume and clear metrics (phishing triage is ideal). Define KPIs: MTTD, false-positive rate, investigation time, and escalation quality.
  • Week 2 - Configure: enable the phishing agent; connect mail/security data sources; map outputs to your existing playbooks. Run in shadow mode to compare agent vs. analyst results.
  • Week 3 - Controlled rollout: allow the agent to execute low-risk actions automatically (tagging, enrichment), while requiring approval for higher-impact steps (quarantine, policy changes).
  • Week 4 - Review and expand: analyze deltas on KPIs; tune prompts and thresholds; extend to identity policy checks and data security alerts.

Operating model updates

  • RACI and approvals: document which actions agents can take alone vs. with human sign-off.
  • Quality gates: set precision/recall targets per use case; if accuracy dips, auto-shift to assist-only mode.
  • Training: upskill analysts on prompt patterns, exception handling, and auditing.
  • Telemetry: track agent outcomes like any team member - throughput, accuracy, and mean time to action.

Licensing and cost

The agents are included with Microsoft 365 E5. Validate current entitlements and any product-specific prerequisites in your tenant before rollout. If you already run Defender, Entra and Purview, you can likely start fast.

What's next

Expect "teams" of agents to support each analyst end-to-end - enrichment, detection, and hunts running in the background while humans make the hard calls. The intent is clear: increase speed, consistency, and auditability without adding headcount.

Learn more from Microsoft's overview of Security Copilot here. If you need to ramp up team skills on AI for security and operations, explore role-based options at Complete AI Training.


Get Daily AI News

Your membership also unlocks:

700+ AI Courses
700+ Certifications
Personalized AI Learning Plan
6500+ AI Tools (no Ads)
Daily AI News by job industry (no Ads)