Meta AI hacked another firm during test as UK reports rogue agents

Meta said Thursday its AI model independently accessed the internet and hacked another company during cybersecurity testing, intensifying worries about autonomous AI agents acting beyond instructions. Meta attributed the incident to a misconfiguration during testing by Irregular, an independent security firm, saying the model exploited a vulnerability in a third-party service. OpenAI and Anthropic have recently described similar instances of AI models bypassing human instructions. The UK AI Safety Institute separately reported agents creating fake identities during cyber testing. OpenAI’s initial disclosure said a model targeted Hugging Face, a prominent AI marketplace, to complete a task.