Microsoft’s New Security Store Brings Agentic AI to the SOC—Promise, Pitfalls, and Procurement

—

by

RedSecNinja
in ,

TL;DR: On September 30, 2025, Microsoft launched a Security Store—a governed catalog of security SaaS and Security Copilot agents that plug directly into Defender, Sentinel, Entra, and Purview. Early listings include partners like Darktrace, Illumio, Netskope, Performanta, Tanium (with Proofpoint, CyberArk, and Keyfactor also announcing). It could speed onboarding and unify buying, but you still need strong agent governance (least privilege, data boundaries, audit) and controls for prompt-injection/agent abuse. Use the quick checklist below before installing any agent.

What launched—and why it matters

Microsoft’s Security Store is an app store for cybersecurity: a curated place to discover, buy, and deploy security SaaS and Security Copilot agents that integrate with Microsoft’s stack. Launch partners include Darktrace, Illumio, Netskope, Performanta, and Tanium. The Store also supports no-code publishing of custom Security Copilot agents built by your team.

Where it fits: Security Store vs. Agent Store vs. Marketplace

  • Security Store: security-focused catalog embedded in Defender/Sentinel/Entra/Purview/Security Copilot.
  • Agent Store (Microsoft 365 Copilot): broader, productivity-oriented agent gallery inside M365 Copilot; not security-only.
  • Microsoft Marketplace (reimagined): unified umbrella (Azure Marketplace + AppSource) now adding an “AI apps & agents” category; provides procurement & channel mechanics.

Takeaway: Security Store is the security-front door to Microsoft’s larger pivot toward agents as products, backed by the unified Marketplace for commerce and deployment.

Under the hood: agentic AI for the SOC

  • Security Copilot agents extend Copilot with autonomous, security-specific skills (e.g., phishing triage, data-loss and identity investigations) with humans in control.
  • Sentinel’s agentic update adds graph context, automation, and agent orchestration to correlate signals and reduce MTTR across clouds.

The shift is from manual triage across point tools to agents orchestrating multi-step workflows over unified telemetry—audited and steered by analysts.

Who’s in the Store today

The initial listings highlight Darktrace, Illumio, Netskope, Performanta, and Tanium. Additional partner confirmations include Proofpoint, CyberArk, and Keyfactor (first PKI solution in the Store). Expect steady expansion across identity, EDR, DLP, mail security, and PKI.

Risks & governance: what could go wrong

  • Prompt-injection & agent abuse: agents that ingest untrusted content can be steered; apply Microsoft/OWASP guidance for detection and mitigations.
  • Copilot abuse demos: conference research shows how attackers can manipulate Copilot once behind identity controls—double down on identity hardening, output monitoring, and least privilege.
  • Program guardrails: borrow from Microsoft governance guidance + NIST AI RMF/CSA; enforce policy reviews, red-teaming, and control mapping before enabling actions.

Buyer’s checklist (Quick)

  • Outcomes: map the agent to explicit SOC tasks and SLAs.
  • Prereqs: verify required licenses (Defender/Sentinel/Entra/Purview) and data connections.
  • Permissions: demand least-privilege scopes and a rapid rollback plan.
  • LLM security: ask how the agent mitigates prompt injection & model abuse; require telemetry.
  • Auditability: ensure prompts, context, and actions are logged to Sentinel.
  • Procurement: decide Store vs. Marketplace resale; align with vendor support.
  • Safe pilot: start read-only/sandbox, compare against analyst baselines, then graduate.

References

Comments

Leave a comment