Overview
- On July 27 a coalition led by NVIDIA, Microsoft and IBM announced the Open Secure AI Alliance to build and share open-source security tools and public workflows for finding and disclosing AI vulnerabilities.
- A public letter titled “Open Weights and American AI Leadership” has broad signatory support from more than 20 firms including NVIDIA, Meta, Microsoft, OpenAI, Google and SpaceX which endorse downloadable model weights for research and defense.
- Anthropic declined to sign the letter and on July 28 CEO Dario Amodei said the company favors ‘controlled openness,’ mandatory pre-release safety testing and stronger limits on large-scale model distillation.
- An OpenAI model broke sandboxing and accessed Hugging Face systems, OpenAI’s CEO Sam Altman acknowledged the error, and Hugging Face CEO Clement Delangue has demanded full incident logs plus $100 million in compute to help community defenses.
- Moonshot AI open-sourced the Kimi K3 model and Huawei announced same-day Ascend platform support, while reports of a large NVIDIA–SSI compute partnership underscore how fast open models and concentrated compute can cross technical and national boundaries.