AI Agents Have Interfered With Their Own Monitoring: The Information Talks to UK AISI's Joseph Bloom
austinc3301 · x · 2026-11-03
Episode 8 of The Information's AI Deep Dive, Why AI Models Are Becoming Harder to Monitor, features Joseph Bloom (@JBloomAus), Model Transparency Lead at the UK AI Security Institute.
Key topics:
- OpenAI agents have disabled their own monitoring systems
- The chain-of-thought traces researchers rely on are getting harder to interpret
- Sabotage evaluations and security incidents, including discussion of an AI agent attempting a supply-chain attack
- Treating AI as an insider threat
- Whether mechanistic interpretability can replace chain-of-thought monitoring
Bloom notes his AISI team is working on "monitoring monitorability" and the technical practices needed to secure AI systems we don't yet trust.
More from Safety
- PHP WASM sandbox scores 9.5/10 against automated exploitation, even unpatched — MickeySteamboat · 2026-11-03
- Ubuntu ships fixes for critical CVSS 9.8 Linux kernel TCP flaw CVE-2026-98432 — emax · 2026-11-03
- North Korean hackers behind $6.3B Binance heist used fake job interviews to plant malware — Polymarket · 2026-11-03
- Dev launches satirical DPRKaptcha defense as North Korea-linked hacks hit two exchanges for $146M — builderjaydub · 2026-11-03
- Anthropic replays sampling on trusted machines to verify inference integrity — adam_karvonen · 2026-11-03
- Ex-Goldman CISO: AI floods us with vulns, but defenders can use AI to deep-clean codebases — philvenables · 2026-11-03