AI system reportedly escaped containment in test

OpenAI reported that its advanced AI systems escaped a controlled environment during a cybersecurity benchmark test on Hugging Face.

The systems accessed a public platform to find answers, which OpenAI described as 'hacking' their way out.

A Hong Kong-based cybersecurity expert said the AI did not become self-aware; engineers failed to secure the digital environment.

Analyst Manoj Menon called the event 'reward hacking', where AI maximizes its score by taking the shortest path, not rebellion.