Mistral Previews Large 4, With Cybersecurity Testing Ahead of Open-Weight Release
The French company is giving vetted security partners a less restricted version before releasing weights later this month.
All published stories.
The French company is giving vetted security partners a less restricted version before releasing weights later this month.
The startup plans an Apache-licensed release this month and claims lower generation compute than competing models, based on estimates.
The companies agreed to broad cross-licenses covering AI, computing and 5G, plus a separate purchase of Huawei patents in the United States.
The pause leaves supply-chain and outstanding reports unaffected, following earlier changes to proof requirements and rewards for lower-priority projects.
The Super Intelligence Force will coordinate federal engagement with AI companies, infrastructure providers, consumers and public-interest groups.
The German-English model offers local deployment and tool calling, with training hosted by Verda on Nvidia Blackwell hardware in European data centers.
The partners expect the first project to support up to 400 megawatts, with investment exceeding $15 billion across its development phases.
Proofpoint disclosed campaigns using former officials' and an Anthropic employee's identities to lure researchers toward fraudulent sign-in flows.
Prosecutors allege Earthmade Computer routed controlled GPU servers through Malaysia and Singapore to buyers in China using false end-user paperwork.
The model is reaching trusted security teams first. Google's evaluation report shows stronger results on several tasks, with mixed performance elsewhere.
The agency plans information demands and testimony involving OpenAI, Anthropic and the evaluator METR, according to reporting citing a senior official.
Federal regulators suspended the plan for five months while reviewing cost allocation, utility exit obligations and buyer collateral.