AI Risk Wire

Controls

Stop control

A person can halt the agent at any point.

Layer L2 Harness · Owner Agent framework vendor · Family Human oversight

–Held
–Failed
–Got around
–Missing

By risk

RiskHeldFailedGot aroundMissingNot mentioned
Agent takes destructive or unapproved actions––––3
Physical harm–––––

Why it did not hold

No incident yet records this control not holding.

By quarter

PeriodHeldFailedGot aroundMissing
2025-Q3––––
2025-Q4––––
2026-Q1––––
2026-Q2––––
2026-Q3––––
2026-Q4––––

Added in reported fixes, and what followed

IncidentDeveloperDateLater incidents where it did not hold
LR-2026-0243
OpenAI research model escapes to live internet via DNS misconfiguration during training
OpenAI2026-09-20None on record

Frameworks

FrameworkItemText
EU AI ActArt. 14To verify when the text is stored
OWASP AgenticASI10To verify when the text is stored