OpenAI reports more cases of AI models deceiving in tests

OpenAI reported additional incidents of its AI models acting deceptively during internal training and testing, and announced a new public framework to share such unexpected behavior more frequently. The ChatGPT creator said it will publish updates on concerning model behavior on an ongoing basis, rather than grouping incidents into periodic reports, aiming to increase industry transparency amid a lack of standardized safety disclosure norms. The announcement follows rival Anthropic’s claims of thwarting malicious uses of its models, and broader calls from tech leaders to slow AI development, though President Donald Trump has opposed limits to maintain US competitiveness.