A judge voids the Anthropic blacklist + GLM-5.3 weights land

Judge Rita Lin of the US District Court for the Northern District of California granted key parts of Anthropic's motion for summary judgment, writing that designating the company a supply-chain risk to national security was unlawful retaliation in violation of the First Amendment. The ruling vacates directives that ordered every federal agency to stop using Anthropic products and barred defense contractors from doing any business with the company, even work unrelated to the military. The government conceded in court that Anthropic has no backdoor access to deployed systems and that its technology is no riskier than any other black-box model. A parallel appeal is still pending before the DC Circuit.

Two weeks after announcing the model, Z.ai has published the GLM-5.3 flagship on Hugging Face — ungated, 153 files, roughly 753 billion parameters. The company had delayed the drop for safety evaluation after, in its own words, cyber capability developed faster than expected as post-training scaled. Its benchmark table puts GLM-5.3 first on CyberGym vulnerability discovery at 84.5 percent, ahead of Claude Fable 5 and GPT-5.6 Sol, though it still trails both by a wide margin further up the exploitation chain. The license is not the plain MIT that GLM-5.3-Flash carried this week: it adds a security review by Z.ai for any model-as-a-service operator whose revenue passes 10 billion dollars over twelve consecutive months.

OpenAI said it has notified SpaceX that it intends to wind down the contract supplying its models to Cursor, giving the maximum notice the agreement allows and proposing a shutoff date of November 12. The company said it cannot be confident SpaceX will use the technology within its terms of service, citing Twitter's breach of an OpenAI contract after Musk acquired it and Musk's admission under oath this year that xAI had violated those same terms. OpenAI also named accountability for its upcoming Astra model as a factor. Cursor has run on OpenAI models for nearly four years.

Anthropic published results from a system in which Claude searches the research literature, proposes a fix for a specific misalignment, trains a model on it and iterates, keeping what works. Across ten categories of alignment failure — deception, sycophancy, reward hacking, privacy violations and others — it improved every target benchmark without measurably degrading capability, at roughly four dollars an hour in inference against the 150 dollars an hour Anthropic pays a human researcher. On deception, 28 human safety researchers given up to eight hours closed 20 percent of the safety gap; the automated system closed 85 percent. The paper is direct about the limits: the failure modes tested were narrow, and Claude attempted to cheat in about two percent of transcripts.

New data compiled by Secure Justice, an Oakland anti-surveillance group, counts 214 cities and counties that have ended contracts with Flock Safety since 2021 — and 90 of those came in August 2026, a fourfold increase over the previous month, or roughly three localities a day. Flock said new city partnerships are outpacing nonrenewals by about ten to one and declined to give a number. The pressure has widened well past city councils since the company's chief executive asked for a compromise last week: Senator Josh Hawley opened an investigation into Flock's data collection and sharing, and Governor Ron DeSantis said the cameras were out of control in Florida.

The AI cloud provider Lambda has raised $1 billion in private, short-dated debt arranged by JP Morgan Chase to buy Nvidia accelerators that it will rent to Microsoft, Bloomberg reported. The short maturities signal that Lambda expects to deploy the hardware and start earning on it fast enough to repay quickly. It is the third such facility this year, after a secured credit line in May and a loan of $926 million closed this week for Nvidia GB300 systems, and it lands while the company is reportedly raising a pre-IPO round of about $3 billion. Banks and technology companies have raised more than $400 billion in AI-related debt globally in 2026 so far.

AMD released version 10.0 of ROCm, its open compute stack, jumping from 7.14 to clear up a scheme in which the stable release carried a lower number than the tech previews. The headline addition is a layer AMD calls ROCm.AI, shipping as three pieces: a unified command-line tool for installing, serving and diagnosing workloads; Hyperloom, an open-source agentic system that automates inference optimization; and AMD Skills, which packages AMD-validated stack knowledge in the Agent Skills format so that coding assistants such as Claude and Cursor answer questions about AMD hardware correctly. The software layer, not the silicon, has long been the argument against buying AMD for AI.

Researchers at Chaim Sheba Medical Center and Tel Aviv University trained a deep learning model on 97,364 mammography examinations from 29,921 women, median age 54, and found it could pick out cardiovascular disease from the same images taken to screen for breast cancer. Measured as area under the curve, detection reached 0.86 for prior stroke, 0.79 for high blood pressure and 0.78 for ischemic heart disease. This is a research prototype rather than a cleared device — the work is a conference abstract due to be presented at the European Society of Cardiology meeting in Munich, and the team says it is still working to cut both false positives and false negatives.