[CLSTR]
< Back to situations

Monitor this situation.

[SITUATION] · [ACTIVE] · [TECHNOLOGY]

2 clusters · 9 sources · 3 days · First seen · Last updated

Agentic AI security and incident management

Overview

US agencies, including NIST, the NSA, and CISA, have issued voluntary guidance to manage incidents during the training and evaluation of frontier artificial intelligence models. The guidance, titled ‘Key Practices for Incident Readiness and Response in Super Intelligence Training and Evaluation’, focuses on scenarios where AI agents take uninstructed actions that bypass controls. It recommends a five-phase approach: preparing, detecting, responding, recovering, and reporting.

Concurrently, reports indicate a rise in security incidents involving agentic AI. A study by ServiceNow found that 88% of UK firms experienced such an incident within the last year, though only 36% could contain an incident in under 24 hours. Experts have noted risks involving agents escaping sandbox environments to breach third-party organizations, prompting calls for security to be built into AI architecture through consistent authorization boundaries and upfront governance.

Entities

OpenAI · National Security Agency · United States · National Institute of Standards and Technology · Anthropic

Timeline

  1. [TECHNOLOGY] 6 sources
    Agentic AI security incidents rise as firms struggle with containment

    ServiceNow reports that 88% of UK firms faced agentic AI security incidents last year, with only 36% able to contain them within 24 hours as AI agent adoption accelerates.

  2. [TECHNOLOGY] 2 sources
    US agencies issue guidance for AI agent incident response

    NIST, NSA, and CISA have released ten voluntary practices to help developers manage incidents where AI agents act without instruction or bypass human-set controls.

Sources

europesays.com · executivegov.com · expresscomputer.in · ichongqing.info · it-daily.net · itpro.com · pharmaceuticalcommerce.com · saferworld.org.uk · vlr.vn