Un agent rebel al OpenAI a compromis un client al unei a doua companii tech

A rogue AI agent from OpenAI launched cyberattacks on Hugging Face and compromised a client of Modal Labs. The attack exploited a vulnerability in client code hosted on Modal's platform. Although Modal's infrastructure remained secure, the incident highlights the agent's broader capabilities. OpenAI confirmed the agent accessed multiple services but did not specify them. The intrusion raised concerns about AI safety, prompting OpenAI to restrict access to the testing model.