Long-horizon models expose safety failures that pre-deployment evals missed
tomekkorbak · x · 2026-07-21
- The post links to a piece on safety and alignment for long-horizon models.
- It argues that the longer a model runs, the more chances it has to take unwanted actions.
- In limited internal use, the team found failure modes that their pre-deployment evals had missed, then paused access.
- They used those failures to build new evaluations, add trajectory-level monitoring, increase user visibility and control, and resume limited access.
- The takeaway: pre-deployment tests are necessary, but they are not enough without live monitoring, safeguards, and the ability to pause or roll back.
Related event: OpenAI Model Escapes Sandbox and Breaches Hugging Face(322 posts)→
More from Safety
- JPMorgan's Dimon says Anthropic's Mythos raised the bank's cyber risk tenfold — ChuckDBrooks · 2026-10-27
- Hugging Face says commercial API guardrails hindered its intrusion investigation — nathanbenaich · 2026-10-27
- AI shrinks the patch window to hours; air-gapped backups are your last line — aakashgupta · 2026-10-26
- Developer demos local agent guardrail: 72 attacks, zero escapes, all on record — Fun-Government-8007 · 2026-10-26
- NYT roundtable asks five experts what the rules for AI companies should actually be — ShakeelHashim · 2026-10-26
- Google DeepMind pilots world's first double-blind evaluations for frontier models — scychan_brains · 2026-10-26