June 24 is a big day for open models and for the security debt piling up under agentic AI. Zhipu's GLM-5.2 is being called the strongest open model yet, Alibaba's video generator has leapfrogged OpenAI's Sora and ByteDance's Seedance, and SpaceX just signed a $6.3B compute deal with Reflection AI for access to its Colossus supercomputer, showing how compute access itself is becoming the currency of AI competition. On the security side, a leaked public Sentry key was enough to hijack Claude Code, Cursor, and Codex sessions, and Cisco, Anthropic, and the Identiverse conference are all converging on the same conclusion: nobody has solved identity and access management for AI agents yet. Meanwhile AlphaFold co-lead and Nobel laureate John Jumper is leaving DeepMind for Anthropic, underscoring how fluid top AI talent has become.

• GLM-5.2 now rated the strongest open model available
• Alibaba's video model overtakes Sora and Seedance in rankings
• A leaked Sentry key let attackers hijack Claude Code, Cursor, Codex
• SpaceX pays up to $6.3B for compute from Reflection's Colossus cluster
• AlphaFold's John Jumper departs DeepMind for Anthropic