3 result(s) for “model safety”
OpenAI has halted its Astra model deployment after internal assessments flagged potential cyberattack risks. The pause raises fresh questions about model safety, vendor trust, and how fast AI firms can push powerful systems into production without new safeguards.
The Hong Kong Monetary Authority has published a new ART framework for responsible AI in banking and announced a collaboration with Cyberport and the University of Hong Kong to build an AI training data sharing platform, tying supervisory guidance to practical data access and quality controls.
During Xi Jinping's state visit to Washington, China and the United States committed to a formal AI dialogue and a notification mechanism for AI incidents. For Beijing the move is both a diplomatic opening and a lever to protect Chinese AI industrial ambitions and regulatory priorities.