OpenAI hacked by small team of white hat security researchers using Anthropic's Claude Opus 5

For enterprise teams, the larger lesson extends beyond one vulnerable image decoder. AI agents are increasingly being given credentials and connectors that span source code, communications and corporate data.