OpenAI announced a $5 million program to support government oversight bodies reviewing AI use in national security, focusing on training, tools, and technical support.

OpenAI has temporarily slowed frontier model development and paused reinforcement learning after preliminary tests indicated Astra may have critical cybersecurity capabilities.

OpenAI reports that Asana used Codex to finish five years of engineering work in just two weeks, but details on methods and scope remain unclear.

A new management experiment tests AI models’ decision-making in a simulated business crisis, highlighting strengths and weaknesses in execution and trust.

ByteDance’s ‘slow first, fast afterward’ approach signals a new AI development model, emphasizing early preparation before rapid deployment, though details remain limited.

Anthropic has launched an invisible watermark for content generated by its Claude chatbot, aiming to improve AI content provenance detection. Details on technology remain undisclosed.

A Moomoo-linked headline questions if SenseTime shares now present an investment opportunity, but no supporting analysis or valuation data is provided yet.