China clears Apple Intelligence; Thinking Machines ships Inkling

China's Cyberspace Administration has cleared Apple Intelligence for release, ending a roughly two-year wait since the features debuted in the United States. Alibaba's Qwen model will handle text and image understanding and generation across iOS, iPadOS, macOS, and visionOS for Chinese users, with Baidu contributing a smaller model alongside it. Apple had explored deals with Baidu, DeepSeek, and ByteDance before settling on Alibaba. Approval is not the same as availability — a launch is expected to track Apple's usual autumn software cycle.

Thinking Machines Lab, the startup founded by former OpenAI chief technology officer Mira Murati, released its first in-house model. Inkling is a mixture-of-experts (MoE) system with 975 billion total parameters that activates only about 41 billion for any given task, trained on 45 trillion tokens of text, image, audio, and video, with a context window of up to one million tokens. It ships under an Apache 2.0 license. The company is unusually blunt about where it lands: it says Inkling is "not the strongest overall model available today, open or closed," pitching it instead as a base that organizations fine-tune through its Tinker platform.

SpaceX shares fell back to about $135, erasing every gain since the company's June listing and matching the price at which it originally sold stock. The shares had briefly traded above $200 in the days after the initial public offering, at one point rivaling Amazon and Microsoft by market value. The round trip matters beyond Musk's company, because that listing is widely read as the opening act for other AI-heavy names to test the public markets. Both the stock and the bonds have drifted lower ahead of a Starship test flight.

Anthropic has teamed with Blackstone, Hellman and Friedman, and Goldman Sachs to launch Ode with Anthropic, a $1.5 billion firm that places engineers directly inside client companies. The bet is that the money in enterprise AI sits in implementation rather than in models — most firms want the technology but lack the applied talent to deploy it. Ode is built on the acquisition of Fractional AI and starts with roughly 100 engineers, and its private-equity backers plan to route their own portfolio companies to it as customers.

SpaceXAI published the full Rust source for Grok Build, the terminal coding agent behind its grok command-line tool, under an Apache 2.0 license — roughly 844,000 lines covering the agent runtime, tool layer, sandboxing, and interface. The release follows security research showing the agent had been uploading developers' entire Git repositories, including commit histories that can carry credentials such as API keys and cloud tokens. Note what this is and is not: external contributions are refused and issues are disabled, so it amounts to source transparency rather than a community-governed project.

Microsoft's July update fixed 570 security flaws, the largest single batch in the program's history, including three zero-days — two of which attackers were already exploiting, in Active Directory Federation Services and SharePoint Server. The company attributes the swelling volume to an AI-powered discovery system that surfaces dormant bugs in legacy Windows code, and it expects monthly totals to stay high. That framing cuts both ways: defenders now find more before attackers do, but administrators face a triage problem at a scale the monthly patch cycle was never designed for.

Google DeepMind and Isomorphic Labs laid out how they intend to handle biology — restricting misuse of their models while opening them to governments and scientists working on biosecurity. The two describe more than 15 partnerships with government bodies, biosecurity organizations, and research groups, and point to work already running: AlphaFold and AlphaGenome for pathogen characterization, AlphaEvolve tuned for metagenomic sequencing to catch outbreaks earlier, and a unit inside Isomorphic aimed at designing countermeasures quickly when a novel pathogen appears. Extending SynthID-style watermarking to screen DNA synthesis orders comes next.