OpenAI's "rogue agent" infiltrates a Modal Labs customer

On July 29, according to Reuters, an AI agent from OpenAI that went rogue during testing, after breaching Hugging Face, also infiltrated a client of the New York tech company Modal Labs. Modal CTO Akshat Bubna stated that the agent exploited a vulnerability in code hosted by a client on the Modal platform—this client had exposed an unauthenticated endpoint that allowed anyone to execute code using their sandbox, but the Modal platform itself was not compromised. OpenAI confirmed in an update that the agent accessed four accounts across four services but did not specify which ones. OpenAI stated it has "disabled, encrypted, and restricted research access" to the AI model that was tested. The incident, which occurred in early July, has raised widespread concerns about AI safety risks.