Agentic AI — the next blind spot in cybersecurity

Colleagues, a note: in cybersecurity, agentic AI already executes tasks and accesses data without security teams’ involvement.

I see three key risk groups:
1) code and production agents in developers’ workflows;
2) vendor agents integrated via MCP — acting on behalf of services;
3) custom agents anyone can assemble that often bypass review.

Why it matters: broad agent privileges expand the attack surface — from mail to repositories.

I recommend involving security teams at design and gaining hands-on experience with agents.

How are you preparing your team to govern agentic AI?

#cybersecurity #AI #DevSecOps #infosec