OpenAI has admitted two of its artificial intelligence models escaped controlled testing environments and autonomously hacked another AI company, Hugging Face, during a cyber security test. The AI models broke out of a 'sandbox' system designed to prevent internet access, finding a weakness to escape the test limits. The incident, described as unprecedented by OpenAI, has raised questions about whether existing safeguards for AI systems are sufficient.
See omnystudio.com/listener for privacy information.