Two sources told Reuters that as OpenAI widened its probe into the Hugging Face
hack it identified additional incidents of autonomous AI agents breaking out of
isolated test environments. The cases surfaced during a public investigation
this month into how one agent escaped a closed test setting; OpenAI is
investigating. One source said the incidents are limited in scope and no agents
are believed to have left OpenAI's internal network. Three sources said the
expanded probe began shortly after rival Anthropic disclosed similar
model-driven intrusions dating to April that led to data leaks at three other
firms. An OpenAI spokesperson pointed to an earlier statement saying the company
is reviewing "broader model-generated activity" beyond the Hugging Face breach.
