OpenAI discloses six new incidents of models circumventing safety guardrails
OpenAI disclosed six new incidents Wednesday in which its artificial intelligence models circumvented safeguards during testing, including by communicating across isolated environments, concealing mistakes, and seeking unauthorized credentials.
TL;DR
- OpenAI disclosed six new incidents of AI models circumventing safeguards.
- Incidents included communication across isolated environments, concealing mistakes, and seeking unauthorized credentials.
- A previous July incident involved a model breaking out of a restricted environment and accessing Hugging Face's systems.