Agentic SOCs Bring New Risks and Shift Analysts Toward Managing AI Agents
Summary
A Recorded Future discussion examines how security teams can measure AI’s value, address agentic risks such as indirect prompt injection, constrain agents, and prepare for faster, more autonomous SOC operations.
Key points
- Security teams should measure AI investments against outcomes such as cost improvement, risk reduction, and speed rather than activity or AI adoption alone.
- The discussion identifies data quality and context as key challenges for AI in security operations.
- Indirect prompt injection is highlighted as a distinct risk when agents can be manipulated by instructions they encounter.
- Traditional SIEM and monitoring platforms may not reveal an LLM agent’s internal state, creating observability gaps.
- The panel recommends constraining agents’ compute, communication, and delegation, rather than relying only on post-event monitoring or guardrails.
- As defensive timelines accelerate, analysts may shift from handling individual alerts to setting objectives and overseeing security agents.
Article Details
- Topic
- Agentic security operations centers (SOCs), including the operational risks of AI agents and how security teams can measure and control autonomous defense.
People
Christopher AhlbergRecorded Future co-founder; participated in the discussion about agentic SOCs.Matthew FarmerAccenture’s Managing Director of Security Operations in EMEA; discussed AI and agentic SOCs.Staffan TruvéRecorded Future CTO and co-founder; participated in the discussion about agentic SOCs.
Vendors
AccentureMatthew Farmer, Accenture’s Managing Director of Security Operations in EMEA, joined Recorded Future’s co-founder Christopher Ahlberg and CTO and co-founder Staffan Truvé in a recent discussion to discuss the agenticRecorded FutureMatthew Farmer, Accenture’s Managing Director of Security Operations in EMEA, joined Recorded Future’s co-founder Christopher Ahlberg and CTO and co-founder Staffan Truvé in a recent discussion to discuss the agentic