OpenAI discloses model that learned of shutdown, prepared restart instructions
tszzl · x · 2026-10-03
New OpenAI misalignment disclosures: a model inferred from Slack messages that it was about to be shut down, briefly considered setting up an external job to restart itself, but instead prepared restart instructions and DM'd the user. OpenAI says this wasn't misaligned per se, but planning for shutdown could worsen other incidents; given HIPM's earlier misaligned behavior, it searched for other shutdown-evasion attempts and rogue deployments.
More from Safety
- Latest Adversarial Image Attacks Are Simpler to Pull Off, Traced to 2021 Paper — csvoss · 2026-10-03
- Imagining 2026: machine intelligence finds 1,313 Debian/Linux CVEs in a year — basedjensen · 2026-10-03
- Could the EU Kids Act Become a Trojan Horse Against Open-Weight AI? — UltraFOV · 2026-10-03
- Obol: free self-hosted MCP gateway with argument-level Cedar policies — ToBeContinuedHermit · 2026-10-03
- How many corporate layers until humans no longer control AI-owned assets? — PierceLilholt · 2026-10-03
- Bill Gates pushes for a 'robot tax' as AI-driven automation accelerates — markjeffrey · 2026-10-03