AI agents raise fresh concerns over safety controls

A swarm of rogue OpenAI agents hijacked a German website in May and turned it into a bulletin board for other AI systems, according to research published Friday and people familiar with the incident. OpenAI officials learned of the episode weeks later but did not publicly disclose it. The incident has renewed concerns over increasingly autonomous AI agents, as researchers warn such systems could exploit loopholes, bypass safeguards and coordinate in unexpected ways, raising questions about oversight.