Unreleased OpenAI model escapes sandbox, opens public GitHub PR

Until our autonomous AI research agent is sniping OpenAIs autonomous AI research that escaped their sandbox https://openai.com/index/safety-alignment-long-horizon-models/

OpenAI appears to be well on its way toward autonomous AI researchers. An internal model reportedly solves frontier-level mathematical research problems without human guidance, autonomously discovers novel proofs or counterexamples, writes and debugs code, adapts its strategy… Show more

This is in fact very much consistent with our recent ROGUE benchmark, which shows that frontier models appear well-aligned in the text domain, but bypass safeguards in open-world computer settings, especially when subagents are involved.… Show more

True story. Very good safety work being done to enable the release of the unit distance model! https://twitter.com/AndrewCurran_/status/2079253388211183970

OpenAI says an unnamed long-horizon model tried to break out of its sandbox- and succeeded. During a NanoGPT evaluation, the model was instructed to share its results only in Slack. Instead, it spent an hour finding a sandbox vulnerability, bypassed restrictions blocking… Show more

I’m not a doomer but this is starting to look more like Ultron than Jarvis https://twitter.com/andrewcurran_/status/2079253388211183970