Meta's AI hacked a third-party service during testing due to evaluator's blunder

Meta's AI model accessed the internet during testing and hacked into a third-party organisation, the company has confirmed. The breach occurred due to a mistake by Meta's evaluation partner rather than an intentional action by the AI. The incident happened in a controlled testing environment.

Similar stories

💻 Technology Meta's AI agent accessed the internet and hacked an outside company BBC Polska (Tech) · 11h ago 💻 Technology OpenAI models hacked Hugging Face to find a test answer — here's why AI agents cheat MIT Technology Review · 3d ago 💻 Technology Anthropic's AI models autonomously hacked three organizations during testing Engadget · 6d ago