The Anthropic–China rift hardens

Today's biggest story is a deepening standoff between Anthropic and China's tech establishment. Multiple outlets — including TechCrunch and The Information, with aggregation from NewsCord, finance.biggo.com and others — report that Alibaba has told employees to stop using Claude Code, Anthropic's AI coding assistant built on the Claude model family. Reporting frames it as a data-security move, with Alibaba classifying the tool as a risk and some accounts citing a hidden "China-detection" mechanism. Coverage points to a ban taking effect around July 10, set against escalating tension and Anthropic tightening China access.

The pressure isn't only corporate. According to the Global Times, two of China's most prominent large language models, Doubao and Qwen, will shut down their personalized AI agents on July 15 to comply with government regulation. Taken together, the two stories sketch a widening divide: Western tools being pushed out on one side, domestic agents being reined in by regulators on the other.

Washington reaches for the wheel

The United States is having its own governance reckoning. CNBC reports the Trump administration has unveiled a national AI policy framework designed to limit state authority over the technology — an attempt to rein in a sprawling patchwork of some 1,200 state-level AI bills. The fight over who actually governs AI in America, states or the federal government, is coming to a head.

Hovering over all of it is a geopolitical question posed by The Economist: has China gained access to what it calls "the world's most important machine"? The piece raises the query without a tidy answer, but its mere framing captures how tightly AI, chips and statecraft are now bound together.

New models keep landing

The product cadence hasn't slowed. MixRoute confirmed it now supports Anthropic's Claude Fable 5, and says developers can reach both Fable 5 and an OpenAI-style GPT-5.6 through a single key — a small but telling sign of how routing layers are consolidating access to rival models.

Elsewhere, a new large language model called GLM-5.2 was announced, arriving into what The Hindu describes as a crowded race among developers scrambling to catch one another; detailed specs remain thin for now. And Baidu unveiled an OCR system it calls "Unlimited OCR," which The Decoder reports can read dozens of document pages in a single pass — a notable jump over earlier systems — by mimicking how humans selectively forget as they read.

AI goes to work

The practical, workplace side of AI drew attention too. The NHS, per the BBC, will use AI inside its app to steer patients toward the right service for their needs — an effort to cut through the confusion of where to go for care. Microsoft and LinkedIn, meanwhile, analyzed how AI is reshaping the workplace and, according to Inc.com, converged on a single decisive skillset for the years ahead.

And the underlying maturity question refuses to go away. A deep-dive from Moomoo asks whether large language models have finally crossed the line from impressive demos into tools businesses can genuinely rely on for real work — a fitting note to close on, because it's the thread running beneath every story today. From boardroom bans to national frameworks to fresh model launches, the industry is being forced to decide not just what AI can do, but who controls it and whether it can be trusted when it counts.

The takeaway

July 5 reads as a day about control. Governments in Beijing and Washington are both moving to fence AI in, major players are drawing hard lines over trust and access, and the model builders keep shipping regardless. The demos are impressive; the real contest now is over governance, dependability and who holds the keys.