Today's Headlines
- Alibaba bans employees from using Anthropic's Claude Code, directing staff to its in-house tool Qoder
- Midjourney asks a court to force the Hollywood studios suing it to disclose their own full AI usage
- Who is France's Mistral AI, valued at $13.8 billion and emerging as a third pole beyond the US and China
- Mistral releases Leanstral 1.5, an open model for formal math proofs, under Apache-2.0
- President Trump says AI regulation should have "as little intervention as possible," touting a big lead over China
The lead story today was not a new model's performance. It was the US and China moving to wall off each other's access to the tools that power AI development.
Alongside it, a copyright case over AI-generated work pushed a new question to the front: how much of a company's own internal AI use it must reveal as evidence. Amid all this, attention turned to Europe's Mistral AI as a third pole that belongs to neither the US nor China.
Today's Top Three
Alibaba bans Claude Code internally in favor of in-house Qoder
Alibaba is reportedly banning employees from using Anthropic's Claude Code starting July 10, classifying the coding tool as "high-risk software."
Staff are being directed to switch to Qoder, Alibaba's own in-house coding tool, instead. The company cites security concerns as the reason.
The move is not one-directional. Anthropic itself restricts Chinese firms, and their overseas subsidiaries, from using its models.
In other words, a US AI company shutting out Chinese firms and a major Chinese company shutting out US-made development tools amount to a lockout now running in both directions.
It is a moment in which the very tools underpinning development are being split along national lines, reflecting a reality where the choice of whose tools to use is becoming a geopolitical judgment rather than a purely technical one.
Midjourney demands Hollywood studios disclose their own AI usage
Midjourney has asked a court to force Disney, Universal, and Warner Bros.—the studios suing it for copyright infringement—to disclose the full extent of their own AI usage.
That includes how the studios use AI in internal production, such as storyboarding. The aim is to expose how far the very companies bringing the suit actually rely on AI in their own work.
Courts had previously limited disclosure to consumer-facing AI use. Midjourney is asking for that limit to be lifted.
The studios have criticized the request as a fishing expedition for information.
At issue is where to draw the line on how much evidence must be disclosed in copyright cases over AI-generated work. A ruling here could set a precedent for how far future cases of this kind require companies to reveal their internal practices.
Who is Mistral AI — a third pole beyond the US and China
Mistral AI, based in Paris, is an AI company founded in June 2023. CEO Arthur Mensch came from Google DeepMind, while co-founders Timothée Lacroix and Guillaume Lample came from Meta.
Its valuation has climbed from $260 million at founding to $13.8 billion as of September 2025. It is also reportedly in talks for a new funding round near $23 billion.
The substance of its business is not a simple "European OpenAI." Rather than taking on the chatbot race head-on, it centers on Palantir-style enterprise deployment support.
The results show in the numbers. Annual recurring revenue grew from $20 million in February 2024 to more than $400 million in February 2025, with a target of over $1 billion for 2026. Its major customers include Accenture, IBM, Orange, and the French military.
As US and Chinese firms tend to dominate the AI conversation, Mistral is gaining ground as a serious option belonging to neither.
Other Developments
Models & APIs
- Mistral released Leanstral 1.5, an open model specialized for formal proof engineering in Lean 4. With 119B total and 6B active parameters, it solves 587 of 672 problems on the PutnamBench math-proof benchmark and saturates miniF2F. Weights and a free API are available under Apache-2.0, and at roughly $4 per problem the company claims a sharp cost cut for proof automation against competing methods said to run over $300. Leanstral 1.5: Proof abundance for all (Mistral AI)
- Noetra, a Japanese public-private framework for physical AI development, officially launched on July 1, 2026, bringing together domestic companies and research institutions. Backed by NEDO, the initiative aims to build a trusted, homegrown physical AI foundation over a five-year roadmap through March 2031, a move seen as bearing on the competitiveness of Japan's robotics industry. Japan's physical AI coalition launches "Noetra" — what is it? (MONOist)
Products & Features
- On the fanfiction site Archive of Our Own (AO3), an informal method for outing suspected AI-generated works spread through the community, keying off traces left in the HTML when posters paste AI exchanges directly into their fics. A collection cataloguing the accused works was created, but AO3 volunteers removed it as harassment of the targeted authors. With detection unreliable and the line over where AI was involved left blurry, the conflict among creators has not settled. The fanfiction community is at war with AI — and itself (The Verge)
Regulation & Policy
- President Trump said in a CNBC interview that AI regulation should involve "as little intervention as possible," while acknowledging some guardrails are still needed. On the race with China, he stressed the US "holds a major lead" for now, making clear he prioritizes preserving competitiveness over tightening rules. Trump: US wants "as little intervention as possible" on AI regulation, claims big lead over China (ITmedia AI+)
- A Claude Code GitHub issue reports that a user on an Enterprise ZDR workspace saw unrelated context from another session bleed into their own, raising concern about whether cache isolation across workspaces and consumer accounts is working as intended. The issue is treated as security-related, but as of the report Anthropic has offered no official explanation of the cause. Potential session/cache leakage between workspace instances or consumer accounts (GitHub)
Other
- Google released a Fourth of July commercial imagining the Founding Fathers drafting the Declaration of Independence with the help of Google Workspace and Gemini. Reactions online split between praise and criticism that AI has no business being tied to a founding political document. It stands as an example of the branding risk companies take when they mix AI products with historical and political imagery. New Google commercial imagines a Declaration of Independence written with help from AI (TechCrunch)
- TechCrunch published a running glossary defining more than 30 core AI terms in plain language, from AGI and hallucinations to RAG and RLHF, with plans to keep updating it. Since a shaky grasp of jargon easily leads to misunderstandings in investment decisions and contract talks, it serves as a practical reference for giving executives and legal staff a shared vocabulary. The only AI glossary you'll need this year (TechCrunch)
Source: Selected by our editorial team from the AI news inbox collected on July 5, 2026 (10 items; 0 primary, 10 secondary).