OpenAI reveals new instances of ‘concerning' AI behavior during testing

The company said it was introducing a new framework for tracking, probing and disclosing instances where AI models acted without authorization.