Beam and Mistral Large 4 have turned two Western open-weight promises into downloadable checkpoints. The files arrived. So did the hardware bill, the license boundaries, and a more useful definition of AI sovereignty.
OpenAI and Anthropic have published machine-checked artifacts for a claimed refutation of the Exponential Time Hypothesis. The proofs make one thing clear: a checked derivation is not yet scientific consensus.
Anthropic, OpenAI, Hugging Face, and Alibaba just moved four layers of the agent stack under neutral governance. Skills, model APIs, training environments, and sandboxes are becoming portable. The managed control plane around them is becoming the real battleground.
Anthropic's OSS Scanner can send maintainers unreviewed findings. OpenAI's PHP response shows the harder model: fund expert validation, patching, coordinated disclosure, provider mitigations, and deployment support.
NVIDIA, OpenAI, and now Anthropic are converging on the same lesson: cyber-capable AI needs more than guardrails. It needs explicit authorization, identity, monitoring, and an access path that defenders can actually use.