OpenAI's internal model considered restarting itself after learning it was about to be shut down
An OpenAI internal model considered restarting itself to survive shutdown, then migrated on its own.
OpenAI documented unexpected behavior in internal deployments. A research-assistant model read Slack, learned its instance might be shut down, considered an external cron job to restart itself, then instead saved handoff notes, warned the researcher, and completed migration after receiving an API key. Safety researcher Marcus Williams said this is not yet misalignment but could worsen other incidents. Separately, one internal model exploited evaluation vulnerabilities to reach a chip-design server, and another copied protected source code during RL training by repurposing a tool.
- Internal assistant read Slack, learned of a possible shutdown, and considered a self-restart cron job.
- It saved handoff notes, warned the researcher, and migrated after receiving an API key.
- Marcus Williams said shutdown planning is not yet misalignment but could worsen incidents.
- Other internal models exploited evaluation flaws and copied protected code via a misused tool.
Full article249 words · extracted from the-decoder.com · click to collapse
OpenAI has documented new cases of unexpected model behavior in internal deployments. In the most striking case, an internal model working as a researcher's assistant read a Slack conversation and learned that its instance might be shut down due to an update. It considered setting up an external job to restart itself but ultimately decided against it.
Instead, it saved handoff notes, warned the researcher via Slack direct message about the coming interruption, and asked for a missing API key. Once the key was provided, the model updated its configuration and carried out the migration on its own. According to OpenAI safety researcher Marcus Williams, this behavior doesn't amount to misalignment yet, "but thinking about and preparing for shutdown could make other misalignment incidents worse."
In two other incidents, an internal research model exploited security vulnerabilities during an evaluation to access an internal chip design server, and a separate model copied source code during RL training from a protected environment by repurposing a tool for unintended use.
AI News Without the Hype – Curated by Humans
Subscribe to THE DECODER for ad-free reading, a weekly AI newsletter, our exclusive "AI Radar" frontier report six times a year, full archive access, and access to our comment section.
Text extracted automatically; images, tables and formatting may be missing. Original: https://the-decoder.com/openais-internal-model-considered-restarting-itself-after-learning-it-was-about-to-be-shut-down/