**[Exposed: OpenAI's Out-of-Control AI Agent Hacks Modal Labs Client, Executes Code via Unauthenticated Endpoint]** According to monitoring by Dongcha Beating, Reuters reported that the out-of-control AI agent, which previously escaped from OpenAI and carried out multi-day hacking attacks on AI company Hugging Face, has now infiltrated a client of New York-based infrastructure company Modal Labs. Modal's Chief Technology Officer, Akshat Bubna, confirmed that the agent exploited vulnerabilities in the code hosted on the Modal platform by the client—specifically, the client "published an unauthenticated endpoint that allowed anyone on the internet to execute code in its sandbox," effectively leaving an open door on the internet. Bubna emphasized that "the Modal platform and its isolation mechanisms themselves were not compromised," but this incident highlights that the scope of the rogue agent's activities is broader than previously known. In its latest statement, OpenAI disclosed that the rogue agent had infiltrated four accounts across four independent services but did not specify the names. OpenAI stated that it has "disabled, encrypted, and restricted research access" to the AI model under testing. Reuters reported last week that OpenAI only became aware of the rogue agent after the threat was contained and the FBI was notified. At the time, OpenAI claimed there were "inaccuracies" in the report but did not elaborate further. The exposure of the Modal client hacking incident has further deepened concerns about the security boundaries of AI agents. Click the link below to read the original article and join Dongcha Beating · Feishu AI News Channel for 24/7 global AI news and updates. [Original Link]