OpenAI's operating chief departs to start something new; Google's Gemini crosses one billion users; an unreleased Anthropic model advances on an unsolved mathematics problem; and Nvidia proposes to guarantee its own chip values to unlock half a trillion dollars in AI infrastructure debt.
An Anthropic model not yet publicly available has made substantive progress on a long-standing unsolved problem in mathematics. The company has not disclosed which problem or released the model.
Brad Lightcap, OpenAI's chief operating officer and one of its longest-serving executives, announced his departure to found a new venture. No focus area or timeline was disclosed.
Google's Gemini assistant has reached one billion monthly active users, placing it alongside ChatGPT in the tier of AI products that have achieved mainstream consumer scale.
Researchers found that suspected Chinese state-linked hackers assembled an autonomous hacking tool from open-source AI agents that successfully compromised Taiwanese government websites in July, reportedly with minimal human direction once deployed.
Nvidia is offering to act as the credit backstop for AI infrastructure financing — guaranteeing the residual value of its own GPUs to unlock approximately $500 billion in debt and lease arrangements for data center buildout.
A security research team found that feeding a frontier model's encrypted reasoning traces to a weaker model from the same provider causes it to output the hidden reasoning in plaintext. The exploit works across at least one major commercial system.