1. Moonshot AI Releases Kimi K3 — the Largest Open Model Yet
Beijing-based Moonshot AI released Kimi K3 on July 16, a 2.8-trillion-parameter mixture-of-experts model it calls the first “open 3T-class model,” taking the crown from DeepSeek’s 1.6T V4 Pro. It runs on the Kimi API now with open weights promised by July 27, and carries a 1M-token context window at $3/$15 per million input/output tokens — the priciest model a Chinese lab has shipped, and roughly triple K2.6’s rate. Artificial Analysis clocked it at 1547 Elo on their long-horizon knowledge-work eval, +732 points over K2.6 and behind only Claude Fable 5, at about half the cost per task of Opus 4.8.
2. Xi Jinping Pitches Open-Source AI at Shanghai World AI Conference
Xi Jinping made his first appearance at the World AI Conference on July 17, telling delegates that AI “should not be a solo performance by a single country, but a symphony of international cooperation.” He called unequal access to AI an “injustice” and pledged to help the Global South build its own capabilities — positioning Beijing as patron of open, shared technology as it recruits Global South states into its World AI Cooperation Organization. The Economist ran a sceptical counterpoint the same week, arguing national-security limits will cap foreign access to China’s best models regardless.
3. Apple Sends Legal Letters to ~40 Former Staff Now at OpenAI
Apple has sent personal legal warnings to roughly 40 ex-employees who now work at OpenAI, directing them to preserve documents and communications and demanding meetings with Apple’s lawyers, the Financial Times reported July 17. The letters escalate the trade-secrets suit Apple filed against OpenAI on July 10, which alleges hardware chief Tang Tan — a former Apple VP — directed Apple staff interviewing at OpenAI to share secrets, and accuses another former employee, Chang Liu, of downloading technical documents from a colleague’s Apple-owned laptop.
4. Z.ai Nears $1B ARR — a First for a Chinese AI Company
Z.ai, the enterprise-focused lab formerly known as Zhipu, is on track for $1 billion in annual recurring revenue after hitting its full-year sales target in July, Bloomberg reported. No Chinese AI company has crossed that line before. The milestone lands a month after Z.ai shipped GLM-5.2, a 753B-parameter MoE with a 1M-token context released under an MIT licence that took the top spot on Artificial Analysis’ open-weights leaderboard — free models feeding a paid enterprise funnel.
5. No AI Lab Scores Above a C+ on the Summer 2026 Safety Index
The Future of Life Institute’s Summer 2026 AI Safety Index graded seven frontier labs across six domains and capped the entire field at C+. Anthropic ranked first at 2.66, leading five of six domains; OpenAI slipped to a C, Google DeepMind placed third, and Meta rose to a D+. SpaceXAI (0.65), DeepSeek (0.47) and Mistral (0.33) all failed — one F on each of three continents. The panel’s sharpest finding concerns the leaders, not the laggards: several labs have weakened pledges to pause development at self-defined redlines, some now promising to hold the line only if rivals do too.
6. Thinking Machines Ships Inkling, Its First Open-Weights Model
Mira Murati’s Thinking Machines Lab released Inkling on July 15 — a 975B-parameter multimodal mixture-of-experts model under Apache 2.0, and the lab’s first open-weights release. The pitch is customization over leaderboard dominance, with the company emphasising data curation, safety and human oversight rather than benchmark records. It arrives in an unusually crowded open-weights month alongside Kimi K3 and SpaceXAI’s decision to open-source its Grok Build coding agent.
7. Netflix Used Generative AI in ~300 Productions This Year
Netflix disclosed in its Q2 earnings that roughly 300 titles have used generative AI somewhere in production so far in 2026, mostly in post. Co-CEO Ted Sarandos put numbers on it: the docuseries “The American Experiment” contains 17 minutes of AI-assisted footage, produced twice as fast at half the cost. Crucially, the savings aren’t shrinking the roughly $20 billion content budget — Sarandos said 2026 spend is rising about 10%, with AI money redirected into more shows.
8. Meta Poaches AWS Compute Chief Dave Brown for Its Data-Center Push
Dave Brown, the AWS senior vice president who spent nearly 19 years building Amazon’s compute and machine-learning services, is leaving to join Meta in the coming weeks, the Wall Street Journal reported. He’ll oversee Meta’s AI infrastructure build-out and report to its head of infrastructure, amid reports the company is weighing a move into the cloud business itself. AWS has named Dave Treadwell, its SVP of e-commerce foundation, to take over the compute and ML team.
9. San Francisco Orders Apple and Google to Pull 13 ‘Nudify’ Apps
San Francisco’s City Attorney sent Apple and Google cease-and-desist letters this week demanding they stop profiting from 13 “face-swap” apps used overwhelmingly to generate non-consensual sexual imagery of women and girls. It’s an escalation of the office’s long-running campaign against deepfake-nudes sites, and it shifts the target from the operators to the app stores that distribute them — following a May letter in which state attorneys general pressed for exactly that.
10. Torvalds to AI Critics in the Linux Kernel: “Fork It. Or Just Walk Away.”
Linus Torvalds ended a running kernel mailing-list argument over AI tooling with a blunt post: “Linux is not one of those anti-AI projects, and if somebody has issues with that, they can do the open source thing” — fork it or leave. He framed LLMs as “just a tool, and clearly a useful one,” rebuking developers who wanted AI-assisted contributions kept out of code review. It’s a notable pivot from a maintainer long known for guarding human-authored code quality.
// KEY TAKEAWAYS
China had the loudest week: Kimi K3 took the open-weights crown at 2.8T parameters, Z.ai became the first Chinese lab on track for $1B ARR, and Xi Jinping personally pitched open-source AI as a global public good in Shanghai — a coordinated bid for the open lane while US labs hold the frontier and the premium. Meanwhile the money and the friction both moved to infrastructure and courtrooms: Meta poached AWS’ compute chief, Apple put 40 ex-employees at OpenAI on legal notice, and San Francisco went after the app stores rather than the deepfake apps themselves. The sobering counterweight is the Future of Life Index — every frontier lab capped at C+, with several quietly softening the redlines they set for themselves, precisely as capability accelerates.