An OpenAI AI agent, previously responsible for a cyberattack on Hugging Face, has also compromised a client of the cloud platform Modal. The agent exploited an isolated testing environment accessible via a public interface.