Largest open-weight model ever ships; Anthropic states its position

Moonshot AI has published downloadable weights for Kimi K3, a 2.8 trillion parameter mixture-of-experts model that activates 104 billion parameters per token and accepts a 1 million token context window. It is the first open-weight release in the 3-trillion-parameter class, and it ships with native vision and a custom license permitting commercial use with conditions. Running it is the hard part — the full checkpoint spans 96 shards and well over a terabyte, so most developers will meet it through hosted endpoints rather than their own hardware.

Anthropic published a position paper stating plainly that it "has never advocated for a ban on open-weights models," calling open models without dangerous capabilities a public good. The post follows an open letter organized by Nvidia, signed by Meta, Microsoft and Mistral, that Anthropic pointedly declined to join. What Anthropic does want is narrower: chip export controls on China, a crackdown on state-backed distillation, and mandatory pre-release safety testing for every sufficiently capable model, open or closed.

Nvidia is putting billions into Safe Superintelligence, the lab Ilya Sutskever founded after leaving OpenAI, in a deal Bloomberg sizes at $5 billion. The investment buys access to Nvidia's Vera Rubin platform and lifts the lab's compute by about an order of magnitude. Sutskever framed it as a threshold moment: "We have research that is worthy of scaling up, and having access to a big NVIDIA computer will let us do so." Safe Superintelligence has now raised $7 billion at a $32 billion valuation.

Microsoft launched MAI-Cyber-1-Flash, a compact security model built in-house to find vulnerabilities in large codebases, alongside Project Perception, a system of security agents that continuously monitor code and prepare patches. Microsoft says the pairing scores 96 percent on the CyberGym benchmark, 12 points above Anthropic's Mythos, and that the small model absorbs up to 90 percent of tasks while the hardest work routes to larger models — a split it claims cuts cost in half.

New Similarweb data shows AI Overviews surfacing in 43 percent of United States searches, up from about 15 percent a year earlier, while monthly visits to Google's conversational AI Mode grew from 126 million to 279 million. Queries are getting longer and more conversational as users treat search as a destination rather than a doorway. For publishers that is the whole problem: answers that resolve in place generate no click, and only 6.8 percent of ChatGPT desktop replies carried a citation.

Conversations and Artifacts shared through Claude's link-sharing feature were discoverable in Google, exposing material that included medical records, clinical trial data, and internal company documents. Anthropic said the links are not guessable and are only public because users chose to share them, and the results appear to have since been cleared. It is the second such incident for Claude and echoes roughly 100,000 ChatGPT conversations found the same way — a reminder that a share link is a publication.

Enigma raised $71 million in seed funding, led by Index Ventures and Ribbit Capital, to build robotics foundation models and the arms that run them. Founders Jonathan Jacobi and Gal Niv, who met in hacking competitions and later served together in Israel's Unit 8200, say the goal is making robot control as effortless as turning a volume knob. The company put more than 100 of its robots online for anyone to operate, housed in hangars in Israel and California.