Anthropic's Claude AI model has identified vulnerabilities in encryption algorithms that are considered difficult to crack, as reported by The New York Times.

DeepSeek AI has published a paper on inference-time scaling for general reward models, introducing SPCT (Self-Principled Critique Tuning). The company hints at its upcoming R2 model. The research addresses reward sparsity and aims to improve scalability of reward models during inference.

Microsoft CEO Satya Nadella warns that businesses relying solely on proprietary AI labs' models risk losing control of their data and thinking, and may not survive. He advises companies to retain metadata to train their own models or use open models, and to keep AI agent infrastructure separate from model providers.