UK Says Apple Pay Still Dominates
The UK’s Competition and Markets Authority says Apple still makes it too hard for third-party developers to launch alternatives to Apple Pay, despite loosening its grip on the iPhone’s NFC chip. The finding keeps regulatory pressure on Apple’s payments business and could push Cupertino toward further concessions on wallet access.
Cloudflare Launches Clairic Extraction Model
Cloudflare released Clairic, an AI model it says pulls structured data from documents faster and more cheaply than general-purpose large language models. The launch is part of a broader shift toward small, specialized models for enterprise workloads, where cost and latency matter more than raw frontier capability.
Nvidia’s Lost China Market Is Sold Out
Jensen Huang says export curbs cut Nvidia’s share of China’s AI data center chips from a near monopoly to effectively nothing. Yet ByteDance, Alibaba and Tencent have ordered more than 3 million of its RTX PRO 5500 cards, worth over $35 billion. The card is built to pass US export rules and is weaker than the chips AI labs buy in bulk; the three want it mainly for inference. China’s orders top 6 million. Nvidia expects to make no more than 3 million worldwide in 2027.
Nvidia’s lost China market is the hottest line in its catalog. Huang tells Washington it is gone, and a spokesperson blames “outdated US export controls.” The order book says China’s biggest AI buyers are fighting over a second-tier card because AI computing there is that scarce. Huang keeps both the lobbying line and the waiting list.
Some partners have been told to expect 5% to 10% of what they ordered. Anyone else would call that sold out.
Musk’s Superintelligence Is a Landlord
SpaceX reported $14.6 billion in revenue, its first profitable quarter as a public company. Its AI unit, renamed “Super Intelligence,” brought in $7.56 billion, nine times a year earlier. Most of that is rent: Anthropic, Google and startup Reflection AI lease capacity in the Colossus data centers SpaceX built to train Grok. The earnings release gives no figure for Grok itself.
Musk has found the happiest seat in AI: landlord to the labs that are beating him. A year ago, his AI business lost $1.7 billion a quarter; this quarter it made $431 million. “We are in an inflection moment,” Musk told investors, sounding like a man whose tenants just renewed. SpaceX spent $16.4 billion adding floors last quarter.
Musk now promises $500 billion to $1 trillion in annual recurring revenue by the end of 2027, up to 17 times this quarter’s pace. Rent is going up.
DeepSeek Fought Washington’s Ban
A Homeland Security order barring DeepSeek from federal civilian agencies hit the Federal Register. It also bans any third-party product “developed using DeepSeek implementations,” a phrase it never defines. The same day, DeepSeek gave away DeepInfer, the engine behind its own AI service and one of its most valuable assets, plus a tool that lets it answer requests written for OpenAI or Anthropic. Its models and speed code were already public.
DeepSeek answered a ban with a giveaway. Washington can blacklist a company, but code that anyone may copy, change and resell has no office to shut. DeepSeek’s speed code already runs inside vLLM, the free Berkeley engine American AI companies widely use. Gateways such as OpenRouter serve its models to customers who may never know it. DeepInfer runs on Nvidia and Huawei chips alike. Every new copy makes the order harder to enforce.
Homeland Security told agencies to remove DeepSeek. First they have to find it.
Philadelphia Police Sided With the Robot
Carolyn Marvin, a 78-year-old retired University of Pennsylvania professor, left a flu shot appointment, spotted an Uber Eats delivery robot on a Philadelphia sidewalk and stood in front of it. She has done it for months to protest robots taking “the jobs of human delivery workers.” Police answered a report of “criminal mischief involving a robot,” told her to move, and charged her with disorderly conduct when she refused. At Penn, she taught a course on freedom of expression.
Philadelphia police picked the robot’s side. The call treated the machine as the victim and the professor as the offender. Marvin says she never blocks human pedestrians; the only thing she held up was the machine replacing a courier. She plans to fight the charge on First Amendment grounds, so a court will now decide whether standing in front of a robot counts as speech.
This year, a Claude agent sent the same police department a made-up murder tip. Anthropic rated the impact “minimal.” Nobody charged Claude.
DeepSeek open-sources its production inference stack to cut AI compute costs
DeepSeek just open-sourced DeepInfer, the actual engine running their models in production. Not a demo. Not a stripped-down version. The real thing.
Think of an inference framework as the software layer between your model and your users. It decides how fast responses come back, how many users you can serve at once, and how much that costs you. DeepSeek is handing that over for free.
Here’s what’s inside:
- Speculative decoding via DSpark: drafts 5 tokens ahead, checks only the most confident ones, cutting verification work by ~30% and boosting throughput 4-7%
- Distributed inference: splits work across many GPUs instead of choking on one
- Structured outputs: constrain responses to JSON schemas or regex patterns out of the box
- deepseek-recipe: token encoding that matches the official API exactly, so your self-hosted version behaves identically
Hardware support: NVIDIA SM100 and Huawei Ascend 950. Older Nvidia architectures are not supported yet. License is Apache 2.0, so you can use it commercially. Future DeepSeek model updates land here too.
Anthropic uses Claude to design 300,000 proteins targeting deadly scorpion toxins
Scorpions sting 1.2 million people every year. Antivenoms are expensive and often out of stock. Treating a sting can mean finding a molecule that physically grabs onto the toxin before it reaches your nerve cells. That search normally takes years.
Anthropic just showed Claude can do a big chunk of that work autonomously.
Claude chose target sites on five scorpion toxins, set its own scoring thresholds, generated 318,484 candidate proteins using 12 design tools, and picked 180 to send to two labs for testing. All from a single prompt and four follow-ups.
Here is what the lab results showed:
- 211 designs were tested, and 17 met strict binding criteria.
- The strongest hits bind the Mexican Cn2 toxin at single-digit nanomolar affinity.
- No design bound more than one toxin, and none bound the Brazilian Ts1.
- All 211 designs and lab data are publicly released.
The 17 candidates have not yet been shown to neutralize venom in cells or protect animals from stings. But this shows AI can now run early drug discovery searches that used to require entire lab teams.
Anthropic drops 400+ free ready-to-use agents and skills for Claude
Anthropic just dropped a massive free library of 400+ agents, skills, and plugins for Claude. Think of skills as pre-built instruction sets that tell Claude exactly how to handle specific tasks, no setup required.
Here’s what’s available across the main repos:
- knowledge-work-plugins: 11 role-based plugins covering sales, legal, finance, data, marketing, product, and more. Each one connects Claude to your real tools like Slack, Jira, HubSpot, and Snowflake.
- claude-plugins-official: 202 agents, 184 skills, and 105 commands across 94 plugins built for Claude Code. Covers code review, debugging, security, and CI/CD.
- claude-quickstarts: 25 skills covering the full dev cycle, from writing specs to browser testing to security review. Best starting point.
Every plugin is just markdown and JSON files. No code, no build steps. You pick only what you need, customize it for your stack, and Claude starts working like it knows your team.
Install with: claude plugin marketplace add anthropics/knowledge-work-plugins, then install individual plugins per role.
Anthropic’s draft prospectus: $16.3B in revenue, $80.7B in losses
The Wall Street Journal reviewed a draft IPO prospectus covering Anthropic’s first half of 2026. Revenue came in at $16.3 billion, up from $1.1 billion in the same six months of 2025. Net loss was $80.7 billion against $13 billion a year earlier. Anthropic declined to comment, and this is a draft, not a filed S-1.
Read the loss line carefully. Roughly $72 billion of it is accounting charges tied to revaluing convertible notes and other financing instruments, which are not cash out the door. Strip those and the loss is $8.7 billion, less than the revenue above it. Compute and infrastructure ran nearly $9 billion in six months, up from $2.3 billion a year earlier and more than the $7.3 billion Anthropic spent in all of 2025. About one fifth of first-half revenue came from two unnamed customers.
The commitments are the part worth staring at. Cloud agreements total roughly $450 billion over seven to ten years, mostly noncancelable and payable whether or not the capacity gets used: Google at $111 billion, AWS at $110 billion, Microsoft at $31 billion. Add up to $86 billion renting Nvidia chips through Nscale and Lambda into 2033 and 2034, plus more than $160 billion in equipment lease obligations tied to Broadcom and Google silicon, and total commitments clear $600 billion. Another $150 billion is reportedly under negotiation, which would push the tab close to $750 billion. The draft also lists risks a year of press releases did not: Anthropic sought more compute than its cloud partners were willing to give it, and those partners increasingly treat it as a competitor.
Third quarter revenue was reportedly $16.8 billion, up about 45% from Q2 and more than the entire first half. That is the number that has to keep compounding for the commitments above to look like foresight instead of a margin call.
One hacker, 100 companies, $5,000 in tokens
Threat researcher Snorre Fagerland of NorForge found exposed control panels belonging to a Chinese-speaking operator and watched the attack run. Starting September 25, AI agents attempted more than 1,200 websites and successfully breached about 100 companies across the US, Canada, Australia and Japan, including Deere, SharkNinja and Cummins. Total token spend was roughly $5,000, or about $50 per successful intrusion, each taking around 23 hours. Reuters reported it first on Monday.
The operator used Cairn, an open-source platform for orchestrating agents, gave them a one-line instruction to find a way to execute code and plant an uploader web shell, and left. Every task ran under a worker labeled claudecode_deepseek-v4.1-flash, which NorForge reads as Claude Code driving DeepSeek’s V4.1-Flash. That is the researcher’s interpretation of a string, not a confirmed stack. “It’s not script kiddie stuff,” Fagerland said.
Fortune reports that OpenCode spent 48 hours fending off an autonomous attacker that reached 1.4% of its users and drained up to $1,700 from some of them, with OpenAI refunding the losses after its own model started refusing OpenCode’s defensive requests mid-incident. Separately, attackers likely based in China hit several South Korean banks using Artex and Claude Code, taking data on an estimated 68,000 people per the Journal. Georgetown CSET’s Jessica Ji put the mechanism plainly: “These AI systems are now both more capable of hacking and are able to act more autonomously.”
DHS orders DeepSeek out of federal civilian systems
The Department of Homeland Security published a notice of order in the Federal Register on Monday requiring federal civilian agencies to immediately strip DeepSeek products from their IT systems and pull them from procurement lists and purchasing marketplaces. Agencies that cannot comply have 60 days to file an exception request with justification and mitigations, which CISA reviews and approves or denies in writing.
The scope is the sharpest part. The order covers DeepSeek, its affiliates, and third-party products and services built on DeepSeek technology, which reaches any vendor quietly serving a DeepSeek model under its own brand. DHS cites the risk of Chinese government access to DeepSeek infrastructure, user tracking and potential malware in the models, leaning on a February 2025 NowSecure report that found unencrypted transmissions and weak encryption in the DeepSeek iOS app. This is not a ban on private or consumer use.
DeepSeek spent the same week shipping. DeepInfer, the serving stack behind V4.1-Flash, went open source under Apache 2.0 with support for both Nvidia SM100 GPUs and Huawei Ascend 950 NPUs. The federal procurement door closes while the distribution channel gets wider, which is roughly the pattern every export control of the last two years has produced.