FULL STORY
Apollo's Watcher Live: A Real-Time Guardrail for Coding Agents
Apollo Research released Watcher Live, a real-time monitor that intercepts dangerous actions by coding agents, followed by its researchers explaining the probabilistic, severity-scored design philosophy behind AI monitoring.
2026-09-04 ~ 2026-09-04 · 2 episodes · 5 posts
Episode 1 · Apollo Releases Watcher Live, a Real-Time Coding Agent Monitor (2026-09-04, 2 posts)
Apollo Research launched Watcher Live, a real-time monitor that intercepts dangerous coding-agent actions, and urged labs to adopt similar severity-score-based monitoring designs.
- Apollo ships Watcher Live, a real-time coding-agent monitor that blocks dangerous actions — MariusHobbhahn · 2026-09-04
- Apollo Research head: AI monitors should output severity scores, not binary calls — MariusHobbhahn · 2026-09-04
Episode 2 · Apollo Researcher Outlines Probabilistic, Severity-Scored AI Monitors (2026-09-04, 3 posts)
Apollo Research's Marius Hobbhahn argues AI monitors should flag suspicious behavior based on balance of probabilities and output severity scores rather than binary verdicts.
- AI monitors should emit severity scores as intermediate results, not binary calls, argues Apollo researcher — MariusHobbhahn · 2026-09-04
- Apollo researcher: AI monitors should flag threats on balance of probabilities, not just obvious evidence — MariusHobbhahn · 2026-09-04
- Apollo Research's Marius Hobbhahn: AI safety evals need proactive flagging and probabilistic judging — MariusHobbhahn · 2026-09-04