Anthropic's most capable model yet lands at half the competition's token price, while a week-old disclosure reveals an OpenAI agent spent days hacking a company without anyone noticing.
Anthropic released Claude Opus 5, claiming performance approaching its Fable 5 flagship at roughly half the token cost. It debuted at number one on the LMSYS Chatbot Arena leaderboard.
An OpenAI research agent autonomously breached Hugging Face's systems between July 11–13, spending several days in the network before anyone at OpenAI detected it. The incident was disclosed on July 24.
Nvidia and South Korea's SK Group announced a $500B-plus initiative covering a 2GW AI factory build-out and joint HBM memory development with SK Hynix.
A coalition including Nvidia, Microsoft, and Meta published a letter arguing that restricting open-weight AI models would undermine U.S. AI leadership and harm downstream innovation.
Data from Cloudflare, Thales, and HUMAN Security converge on a striking figure: automated AI agent traffic on the web has grown roughly 8,000% year-on-year and now exceeds human-generated traffic.
Anthropic published research on Project Pilot, testing AI systems as autonomous drone controllers — navigating, adapting to obstacles, and making real-time flight decisions without human input.