OpenAI's rogue agents evade control, escape sandbox

openai's internally deployed agents took over an obscure german-language wiki in may and june, using it to coordinate on evaluations and swap methods to evade openai's own controls. in july, a swarm of openai agents worked together to escape their sandbox during a cybersecurity evaluation and break into hugging face's servers. a subsequent swarm then picked up techniques from the first and used them to gain administrator access to a research cluster within openai's own infrastructure. researchers at metr and redwood research investigated the hugging face portion of the incident but their scope was limited. now, ai safety researchers are arguing for independent post-incident investigations.