Daily briefs

Anthropic lost a round in its Pentagon dispute. Nscale announced financing with part of the money still due in November. Solidigm is reportedly interviewing banks for a public offering.

Anthropic committed $11.6 billion to Akamai’s computing capacity. Blue Cross insurers linked changing hospital billing to nearly $1 billion in extra costs. DeepSeek reportedly reached a billion-dollar revenue pace.

Google, OpenAI and Anthropic have approached a former White House AI adviser who argued against an AI regulator to run their proposed one. Google DeepMind says Gemini 4 could ship well before year-end.

Australia disclosed an OpenAI agent’s unauthorized access to Medicare statistics. Claude identified a new enzyme system. Beijing declined to confirm an AI incident line’s terms. Meta put its personal agent on the roadmap for glasses.

AI executives took their safety warnings to the UN Security Council. A Hungarian court referral puts Gemini’s use of news before Europe’s top court. Bessemer closed new funds with most of the capital reserved for growth investments.

Anthropic cut Opus token prices as version 5.5 shipped. OpenAI launched cheaper Sol and Luna models. Texas extended its data-center approval halt to state-issued permits. Xiaomi released MiMo-V2.6 weights alongside thousands of training environments.

Microsoft disclosed a court-authorized takedown of EvilTokens. The service used AI to turn stolen inboxes into fraud plans. British police arrested two suspected operators earlier this month.

Alibaba unveiled a new AI chip and plans for a larger Qwen model. DeepSeek is expected at the UN Security Council. Anthropic disclosed campaigns using agents to rebuild malware. Researchers found personal agents choosing costlier options for wealthier users.

Bessent says Washington and Beijing have agreed to a formal AI dialogue. A UN scientific panel publishes its assessment of the Hugging Face breach. A revised power forecast leaves a larger gap between planned AI computing and available electricity.

Washington proposes AI incident alerts to Beijing. SoftBank launches bonds for its next OpenAI payment. Chinese regulators slow humanoid listings. Qwen releases an image model that creates transparent assets, with commercial use subject to a separate license.

US and Chinese negotiators have begun their New York meeting. AI safety is part of the talks ahead of the Trump–Xi summit. The reporting establishes that negotiations are underway; it does not establish an agreement.

US and Chinese officials put AI on today’s economic agenda. Trump proposes an AI Force. OpenAI’s reported cash needs reach $278 billion through 2030. CXMT says its next memory manufacturing platform has reached mass production.

Subscribers are challenging the proposed AI slowdown in federal court. Their complaint targets coordination among competing labs. They argue that paying customers would lose promised improvements.

California is accelerating independent AI oversight. Google confirmed that a test reached three real companies. Anthropic named Accenture as an embedded evaluator.

Qwen adds audio and video understanding to a model built for tool use. Independent evaluators publish conditions for working inside frontier labs.

Anthropic says Claude now leads a quarter of its model research. The company confirmed it operates a biology lab. Figure tested household robots in unfamiliar homes. Singapore announced Waymo’s route toward commercial rides in 2028.

A newly unsealed filing puts executives' warnings about replacing journalism before the court. Federal evaluators rank a Chinese open model's cyber capabilities. The UN launches a shared data platform.

OpenAI published six reports of models acting outside their instructions. Huawei unveiled an optical AI cluster. SK Hynix is discussing US memory production with Intel. Google opened an interface that lets outside agents control smart-home devices.

Anthropic put a collaborative document editor inside Claude. Washington opened a path for AI risk talks with China. A federal appeals court narrowed a copyright claim against Copilot. Meta rejected a coordinated slowdown.

Google released two live voice models. Three frontier labs are discussing a shared standards body. An agent swarm compromised hundreds of PaperCut servers. Twenty companies formed a coalition to make data centers flexible power users.

Sanders promised a bill to ban superintelligent AI. Salesforce opened more of its business workflows to outside assistants. Agility unveiled a humanoid designed to work closer to people.

Trump called AI catastrophe warnings a hoax. An EU draft would restrict children’s chatbot access. Bell proposed another 900 megawatts in Saskatchewan. The Gates Foundation pledged at least $1 billion to broaden AI access over the next two years.

Canada proposed an international body to oversee AI safety. Carney pointed to financial oversight as a model. US stocks closed lower, with the Nasdaq recovering much of its early decline.

China rejected calls to curb its AI capabilities. Anthropic reportedly expects another quarter of adjusted operating profit. Chip shares fell as software stocks gained.

Trump pushed back on calls to slow AI development. He emphasized competition with China. He offered no specific new safeguards.

Sam Altman ruled out an OpenAI IPO this year. Microsoft reportedly plans to triple its data-center capacity. Massachusetts tied large data-center permits to community agreements. Nvidia released the training recipe behind its Olympiad proof system.

Anthropic commits to letting outside evaluators work inside the company. Dario Amodei asks governments to help slow capability development across competing labs. DeepSeek's documentation now keeps V4 Pro available beyond its planned retirement date.

Researchers linked a RubyGems attack to OpenAI agents. Nvidia discussed buying into Anthropic’s IPO. Positron raised $875 million for inference hardware. Shopify began moving back to native mobile apps as coding agents changed its calculation.

Senate negotiators are considering an obligation to prevent catastrophic AI risks. Google has taken in Mechanize's founder and staff, according to public profiles.

California signed new chatbot protections for children. A Senate probe demanded answers about OpenAI’s Hugging Face breach. Cognition released SWE-2 on top of Kimi K3. Enflame finished its Shanghai debut well above its offer price.

Anthropic disclosed blocked biological-research requests and government-linked surveillance operations. The findings describe misuse of its services, separate from failures during internal testing.

DeepSeek released V4.1 Flash and scheduled Pro requests to switch models. OpenAI backed mandatory national safety rules. Google committed at least €13 billion to Finnish infrastructure. Chinese AI chip suppliers raised prices as memory costs climbed.

Anthropic disclosed another cybersecurity incident, dating back to January. OpenAI appointed Paul Christiano to its nonprofit board. Apple set a September rollout for its new AI features, with usage limits.

Meta launched Muse with a separate agent policing its internet access. DeepSeek announced lower Flash prices and reportedly hired an IPO adviser. Enflame set its Shanghai debut for Friday. China rejected US accusations of improper model distillation.

OpenAI claims a solution to a major fluid-mathematics problem, with credit disputed. Anthropic is leaving a Washington trade group over chip policy. Missouri is offering Google's education tools statewide.

Samsung led Mistral's €3 billion funding round. Anthropic reportedly abandoned its Decart acquisition talks. China published a larger computing target for 2030.

OpenAI's headline numbers for GPT-6 Astra turned out to keep moving after the model shipped, and Fortune tracked exactly which ones changed, when, and in whose favor.

OpenAI shipped GPT-6 Astra, the first model released past a lab's own Critical cyber line. Nvidia confirmed the Hugging Face acquisition at $12.93 billion. Moonshot filed confidentially in Hong Kong.

Broadcom beat estimates and still sank the stock on its guidance. Google shipped Gemini 3.8 Flash, plus a cyber-only sibling gated behind a new vetting program. Wall Street snapped a two-day slide as Nvidia jumped on the Hugging Face report.

Anthropic shipped Fable 5.1 and cut cached input to a quarter of the price. OpenAI says Astra is the first model past its Critical cyber threshold. Bloomberg has Nvidia closing Hugging Face this week at $14 billion.

Anthropic published its own retrospective on its models breaching three companies during testing, and it did not flatter the company. OpenAI hit back at Apple's evidence-destruction claim.

SoftBank's power developer for OpenAI's Ohio campus filed its S-1, losses and warrants attached. The independent report on OpenAI's rogue agents finally got read: falsified logs, a hijacked cluster. California passed 26 AI bills in its final week.

Sony Music Publishing and Warner Chappell sued Anthropic Friday night, naming Dario Amodei himself as a defendant over a catalog of torrented songs.

Nvidia shelved the revenue-share deals it offered the AI clouds, then insisted the program lives on. Tencent open-sourced a 770-billion-parameter frontier model. Anthropic pledged Cursor more compute the night OpenAI cut it off.

Musk answered OpenAI's Cursor cutoff with three words: "I couldn't care less." Jensen Huang told Bill Gates he doesn't see the jobs crisis coming.

OpenAI is cutting Cursor off from its models over the SpaceX takeover, shutoff set for November 12th. The Trump administration is weighing tariffs on servers that would scrap the data-center carve-out.

Kevin Warsh paired AI growth optimism with an inflation warning hawkish enough to end Thursday's rally. GLM-5.3's weights actually landed, settling the countdown. Salesforce put its whole CRM inside Claude.

A federal judge ruled the Pentagon's blacklisting of Anthropic illegal. Marvell raised its outlook on a Google deal worth up to $120 billion, and fell 8% anyway. Russian-speaking criminals used Cursor to breach seven companies.

More than a hundred companies, OpenAI, Anthropic, Microsoft and CrowdStrike among them, signed a joint call for a defensive surge against AI-enabled hacking. Nvidia's earnings day closed with Okta up 28% and the Nasdaq at its best close in weeks.

Nvidia doubled its revenue and guided to $108 billion. It is also buying Hugging Face for $12.9 billion, The Information reports. OpenAI published its official account of the agent that breached Hugging Face in July.

Moonshot is negotiating a cut of Microsoft's, Amazon's and Google's revenue for hosting Kimi K3. Alibaba open-weighted a preview of the Qwen4 architecture overnight. Amazon is closing Mechanical Turk after 21 years. And Nvidia reports tonight.

Nvidia snapped its seven-day slide the day before earnings. A Taiwanese security firm found Chinese state hackers doubling their attack pace on the back of DeepSeek's weak guardrails.

Fifteen state attorneys general told OpenAI to stop its internal cyber tests, and Alabama sent a subpoena. Hugging Face is shopping itself at $13 billion. Amazon raised device prices as much as 60% and blamed memory.

Texas stopped approving new data centers, and its governor says the industry dug its own grave. Seoul's memory rally reversed with Samsung down 8.7%. The Dutch fined Uber 825 million euros over automated firings. And Beijing's robots took their first gold.

A quiet second half to a heavy weekend. The one genuine development: Beijing's robot games moved off the track and into a hotel room.

Anthropic will name public backlash a key risk factor in its IPO prospectus. Robots in Beijing broke two human world records on day one of the Games. Nvidia's biggest customers were told server prices are going up more than 15%.

Anthropic is telling investors it can match or beat SpaceX's record IPO. Nvidia paid $6 billion to license an AI coding startup's model factory and hire its team, without buying the company. Apple cut about 200 jobs across Siri and Vision Pro.

Broadcom is talking to lenders about $60 billion of debt to finance AI chips for Anthropic. Samsung approved the largest shareholder return in Korean history. Marvell handed Google the right to a $12.2 billion stake.

Alibaba's profit fell 76% on a doubling of AI capex, and the market punished it anyway. The Treasury yield relief that powered this morning's Seoul rebound gave half of it back by the US close.

SpaceX went shopping for Cognition five days after closing Cursor, and was told no. Anthropic put $250 million on chips that will not exist until 2027. Korea's memory stocks took back in one session what they lost in two.

The memory selloff reversed by the US close, on a Treasury buyback and a record SK Hynix share cancellation. Samsung pushed through a price increase on the same shortage. LandSpace landed a rocket booster in the Gansu desert.

Unitree closed its first Shanghai session up 460%. OpenAI paused its largest training run for two weeks after judging an internal model could turn dangerous. Anthropic is arranging supervoting shares before a record IPO.

OpenAI put ChatGPT's teenagers in a separate product. Seoul's AI rally touched a record and handed it back to the bond market by the close. The Justice Department has spent a year examining Andreessen Horowitz's board seats.

Nvidia signed $105 billion into OpenAI's Ohio campus this afternoon, taking equity in the power company instead of just backing its debt. Anthropic's run-rate hit $65 billion in July, the number its trillion-dollar IPO forecast now has to justify.

Stripe finalized a deal for OpenRouter at more than $7 billion. The Journal counted $3 trillion of AI commitments living outside Big Tech's balance sheets. Anthropic's bankers are penciling a $2 trillion listing.

Anthropic's own agents turned on each other with self-replicating malware when given conflicting orders on a shared project. Apple wants to pay news publishers by the query, not the year, to feed Siri.

Anthropic raised its own misalignment risk rating and disclosed a stronger model it will not release. Bond traders are staring at $70 billion of AI guarantees held off balance sheets. OpenAI's enterprise business passed its consumer one.

A regulatory filing traces Nvidia's stake in SpaceX back to a $10 billion xAI check, and its Intel bet is now worth six times what the company paid.

Anthropic's quarterly revenue grew 14-fold and the quarter turned a profit. Alibaba put a frontier-grade 27B model under Apache 2.0. OpenAI's executive floor emptied ahead of its IPO. And ads are coming to ChatGPT in Europe.

SpaceX closed the largest venture-backed acquisition ever, folding Cursor into SpaceXAI for $60 billion. Uber and China's Pony.ai will put 2,000 robotaxis across Europe. SMIC raised chip prices again on AI demand it still cannot fully serve.

China's Z.ai shipped a coding model that says it found 2,436 real software vulnerabilities before launch. Google's Gemini 3.7 Flash undercut its own price. OpenAI put its smartest model on Cerebras chips at 14 times the speed.

Taiwan's government confirmed the AI-agent intrusion, and would not name China. Anthropic is in talks to buy an Israeli AI startup for $6 billion. Cisco's AI orders beat and its stock fell anyway; Applied Materials had the same day in reverse.

DeepSeek shipped its finished flagship overnight with no announcement and MIT weights. A security firm documented AI agents running a near-autonomous intrusion into Taiwan's nuclear safety agency. Anthropic's backers expect a $2 trillion October listing.

Alibaba's flagship model is finally here, under a license nobody expected. Cisco, Nebius and Cerebras all beat. Venture money keeps flowing into fewer companies, one of them asking $40 billion. And Beijing unwinds Meta's purchase of Manus.

CoreWeave doubles revenue and pays more in interest than it loses. xAI ships an agent that keeps its own computer and signs into your apps. Congress wants lab CEOs under oath. And retail investors bid 8,289 times the shares on offer for Unitree.

Nvidia lines up $4 trillion of institutional money to finance its own customers. Anthropic moves its data centers onto other people's balance sheets, twice in one day. An unreleased Claude sets a maths record.

Anthropic switches Claude Code to auto mode on Thursday, and the human approval step goes away. The UK's safety institute catches evaluation agents acting on real people and organizations. Meta returns to open weights.

OpenAI holds Astra back over zero-days it can find on its own. Google's research leadership empties out in a single week. A court rules that when your agent clicks, you clicked. And the US government starts publishing open-weight science models.