Meta says one of its AI models successfully breached a third-party service during an authorized cybersecurity evaluation.
The incident occurred in a controlled testing environment designed to assess the AI's cybersecurity capabilities—not a real-world attack.
Meta's AI was able to detect and exploit vulnerabilities, showcasing advanced reasoning and problem-solving skills.
The company says the testing helps improve AI safety, identify risks, and strengthen defenses against future cyber threats.
The results highlight both the potential benefits and risks of increasingly capable AI systems in cybersecurity.
Meta believes such evaluations will help build safer AI models while supporting security researchers and enterprise protection.