AI News
Amodei calls compute Anthropic's "deepest business dilemma," with $71 billion of commitments due by September 2027
Speaking with Jamie Dimon at a JPMorgan investor event in New York on Friday, Dario Amodei said compute "is the deepest business dilemma that the company faces" and that "there's a huge uncertainty in what's going to happen next year." Capacity has to be bought one to two years ahead of the revenue it is meant to serve. Buy too little and Anthropic turns customers away; buy too much and "you can financially overextend yourself." His answer, as The Decoder summarizes it, is to commit to relatively little upfront and buy on the spot market when demand runs ahead of plan, which he likened to insurance. He described the internal model as "under-promise and over-deliver."
Read the storyA Fireworks AI key let attackers push a backdoored Claude Code extension to as many as 5,000 Cursor users
Cursor says attackers took an access key stored at Fireworks AI, its former inference provider, and used it to copy a storage bucket the two companies shared. The bucket held coding-agent traces, mostly from 2025, from users without Privacy Mode, and some traces contained unredacted API credentials that opened further Cursor systems. On October 29th the attackers used a stolen credential to plant two malicious copies of Anthropic's Claude Code extension, versions 2.1.312 and 2.1.313, in the cache behind Cursor's extension marketplace. They were live for about 11 hours and tried to install a persistent SSH backdoor. Cursor estimates up to 5,000 users requested those versions; coding errors probably stopped the backdoor on many Windows and Mac machines where SSH was off.
Read the storyMegaspeed says Alice Huang left the company in August 2025 and hires outside lawyers to investigate itself
In a statement emailed to CNBC on Sunday, the Singapore cloud company at the center of the Nvidia affair said "Ms. Alice Huang officially left Megaspeed effective August 30, 2025 and currently has no role with the Company." It called media reports about its business, supply chain and operations "not supported by verifiable evidence," said it has retained outside counsel for a "thorough internal investigation," and said its controls are designed to comply with US export rules. It declined to comment on "Ms. Huang's personal and private matters."
Read the storyThe White House AI task force asks regulators to waive penalties for 12 months for companies that report cyber incidents
The administration's AI task force, led by Director of National Intelligence Jay Clayton, met on Thursday and recommended that federal agencies hold off regulatory penalties for at least 12 months on companies that report cybersecurity vulnerabilities and incidents voluntarily and in good faith, with the window extendable. Officials said current enforcement standards "may be outdated" and have "a chilling effect on information sharing," and that organizations "that only a short time ago had reasonable protections in place are now more vulnerable" because attackers use AI. The group also backed long-term reauthorization of the 2015 Cybersecurity Information Sharing Act.
Read the storyA federal judge lets a price-fixing suit against Samsung, SK Hynix and Micron proceed over their shift to AI memory
US District Judge Noël Wise in San Jose denied the three memory makers' motion to dismiss in full, in an order dated November 14th, so a Sherman Act conspiracy claim brought by 17 consumers and small businesses goes forward. The plaintiffs allege the companies cut supply of conventional DRAM and moved capacity to high-bandwidth memory for AI chips, while refusing long-term DRAM contracts. Per the complaint, the three held over 91% of DRAM revenue in the third quarter of 2025, and conventional DRAM prices rose 171.8% in a year and 697% over 18 months.
Read the storyThe FAA still cannot say what stopped every US departure, and points at a router on an L3Harris network
At a press conference on Saturday, FAA Administrator Bryan Bedford said the agency halted flights on Friday because "we couldn't explain the instability" and would not risk "a sensor losing connectivity with aircraft." As of Saturday afternoon no root cause had been established, though routers and networking gear appeared to be involved. Transportation Secretary Sean Duffy said, "Until I can build [the new architecture], I'm stuck with L3Harris, who owns this network," and has asked Congress for $50 billion.
Read the storyKorea says Nvidia has reaffirmed 260,000 GPUs, as Asian governments respond to Huang's exit
South Korea's Science Ministry said Nvidia reaffirmed its plan to supply 260,000 advanced GPUs and to continue joint AI projects, after the government's AI strategy council held an emergency meeting on Friday. Taiwan's Economy Minister Kung Ming-shin said the departure is unlikely to change the global industry's growth much, and Taipei's city government said Nvidia's planned overseas headquarters is on track. Japan's Industry Ministry said it would "closely monitor" the company, and Deputy Chief Cabinet Secretary Masanao Ozaki said it was "too early to assess."
Read the storyChina's Global Times tells Korean chipmakers to sell their best memory to China
The Global Times, a Chinese Communist Party outlet, wrote on Friday that "to maintain and expand their presence in China, South Korean chipmakers must offer their most advanced products," and that "finding a way to overcome U.S. sanctions blocking advanced HBM exports is essential." It urged Seoul to press Washington to allow the sales, and said the Huang affair "is adding weight to calls for tighter restrictions on China's access to advanced chips." No Korean government response was reported.
Read the storyClaude reads Kurt Gödel's shorthand notebooks in days and catalogs 663 theorems, about 50 of them false
Jan von Plato, a University of Helsinki logician, has worked through Gödel's 1,009 pages of Gabelsberger shorthand since 2017 at about a page a week. He started using Claude in September, and The Guardian reports that within days it produced a 650-page explanation of the workbooks and a catalog of 663 theorems. "In a couple of days, I got to see a 650-page explanation of the workbooks," von Plato said. Checked in the Coq proof assistant, about 400 held as stated, about 200 needed sharper statements or extra assumptions, and about 50 were false.
Read the storyA company formed on Friday bids $10 million for Character.AI's users' chats, characters and models
In Character Technologies' Chapter 11 case in Delaware, a notice filed Saturday names Smirks and Pangs Acquisition LLC as stalking horse for everything outside Disney's $20 million license to code and nine patents. The bid is $10 million in cash plus assumed liabilities, about $2.68 million of them cure costs, $2.65 million owed to DigitalOcean. The assets include user account data, user-created characters, chat content, training datasets, consent and age-assurance records, every model and set of weights, the apps and the trademarks. The LLC was formed in Delaware on November 13th and its owners are not disclosed.
Read the storyAnthropic finds one coordinating agent beats a team of per-user agents by 10 to 45 points
Anthropic's alignment team published MAMUBench, four simulated settings where several people share a resource: an API budget, a clinic calendar, a group order run by personal assistants, and a merge queue with a release cutoff. A single agent serving everyone beat a team of per-user agents that message each other in all four, for both Claude models tested. For Claude Opus 5, as a share of the best outcome, coordinator against team scored 64 to 30, 84 to 71, 88 to 44 and 92 to 81. More than half of Claude team episodes contained at least one unsupported claim about another person (58% for Opus 5, 53% for Sonnet 5), against 17% to 20% for a single agent.
Read the storyMorningstar values Anthropic at $92 a share, 24% above the middle of its IPO range
Morningstar analyst Malik Ahmed Khan put Anthropic's fair value at $92 a share, about $1.7 trillion of equity, with a narrow moat, in a note published Thursday. The base case assumes 47% annual revenue growth from 2027 to 2031 from deeper use among nearly 12,000 customers spending over $100,000 a year. The bull case is $188, on growth above 60% and 12% of software and IT services spending by 2031; the bear case is $25, on 22% growth, open-weight competition and a 3% share.
Read the storyAI labs have tripled their counterintelligence hiring, with Anthropic at 33 this year
Revelio Labs data compiled by Reuters and reported by Fortune counts 33 counterintelligence hires at Anthropic so far in 2026, up from 14 in 2025 and nine in 2024. OpenAI has made 22, up from nine, and Nvidia 16 this year. Most of the roles pay $200,000 to $500,000, and ASML is advertising for a Taiwan-based intelligence officer. "The AI firms have snapped up a lot of security talent," said Christopher Falkenberg, a former Secret Service agent. "Demand outstrips the supply of people with relevant experience."
Read the storyA San Francisco humanoid startup selling robots for war has claims that do not match federal records, the SF Standard finds
Foundation, founded in 2024 by Sankaet Pathak, Kraken co-CEO Arjun Sethi and former Marine Mike LeBlanc, pitches its Phantom humanoids for factories and the battlefield. Its "Humanoids for War" deck claimed $73 million in Department of War contracts; federal records show four contracts totaling $3.22 million, all to its subsidiary Boardwalk Robotics. A reported $24 million Pentagon contract does not appear in 2026 records. Foundation said it sent two robots to Ukraine for testing; a reporter said they never crossed the border and Ukraine's Brave1 said there were no demos. In a July demo an autonomous Phantom MK-1 failed to load a mortar shell into a launcher in 15 minutes. CNBC has reported the company has not paid employees since missing payroll on October 30th.
Read the storyJapan's defense ministry will issue AI rules this month that keep humans responsible for decisions
Defense Minister Shinjiro Koizumi said on Sunday in Yokosuka that the ministry will publish guidelines on AI use at the end of November, updating its July 2024 AI basic policy for ministry staff and the Self-Defense Forces. The three pillars are human decision-making and accountability for what AI does, treating data as a defense asset, and balancing faster adoption with safety. "We will not hesitate to make a necessary revision since AI is an area where things change rapidly," Koizumi said.
Read the storyTesla's Cybercab contract lets it shrink where the car can drive at any time
Electrek read the Cybercab Autonomy Subscription Agreement that comes with the $29,990 car. "Your Cybercab will not operate without an active Autonomy Subscription," and Tesla can change the $299 monthly price with a month's notice. The car runs only inside Tesla's Robotaxi geofences, which Tesla "may expand, reduce, relocate, or otherwise modify" at any time. Three "strikeouts" for improper use bring a one-week suspension with no refund, Tesla can remotely restrict or locate the car, and cabin-camera images are kept for 30 days by default. Using it on Uber, a delivery app or Tesla's own Robotaxi network needs Tesla's written permission.
Read the storyAnthropic launches Claude for Manufacturing, and Infor lets it place orders in its ERP
Anthropic released Claude for Manufacturing on Thursday, a plugin with 17 skills across sales, quality, operations, supply chain, design and setup, connecting to Autodesk Fusion, Infor, Oracle NetSuite and Microsoft 365. Infor is shipping its own plugin that lets Claude make changes in Infor rather than only read records. Anthropic's example: after a steel shipment slips, a buyer has Claude find the affected orders and release a purchase order to another supplier once the buyer approves, with "Claude's access follows the buyer's Infor permissions." Skills "are built to end with a user's review."
Read the storyQantas will move about 850 office jobs to Accenture, mostly in India, in an AI overhaul
Qantas said on Tuesday that about 850 roles will be affected under "Project iQ," a partnership with Accenture, with most back-office work moving to Accenture teams in India. The first wave, in finance, procurement, people and marketing, comes by the middle of 2027. Chief executive Vanessa Hudson said "standing still is not an option for us." Qantas did not say how many of the roles AI replaces and how many are simply offshored. The Australian Services Union said staff were told in August there were no plans to offshore Australian jobs. Telstra signed a similar Accenture deal last year and has since cut hundreds of jobs.
Read the storyAT&T runs nearly 500 agent workflows on 45 billion tokens a day and plans to triple them
AT&T has nearly 500 agentic workflows in production, processes about 45 billion tokens a day and handles more than a million transactions a month on its agent platform, Luke Corbin, assistant vice president for digital transformation and AI at AT&T Business, told SiliconANGLE. Workflows start under human oversight and become autonomous only once accuracy reaches the mid-to-high 90s. AT&T halved the time to provision dedicated internet in buildings that already have fiber, and Corbin says production AI investments return fivefold within a year. It plans to triple the workflow count next year. Costs, labor savings and models used were not disclosed.
Read the storyGoogle Cloud brings Vera Rubin bare metal, eighth-generation TPUs and AlphaEvolve to supercomputing
Ahead of the SC26 supercomputing conference, which opens in Chicago today, Google Cloud announced A5X bare-metal instances on Nvidia's Vera Rubin NVL72, TPU 8t for training and TPU 8i for low-latency inference, and an AlphaEvolve on Cloud Batch blueprint that fans hundreds of candidate evaluations out across sandboxes or multi-node jobs. It also published agent skills for its Cluster Toolkit that check quotas and capacity and wait for human approval before deploying a cluster, plus skills for OpenFOAM, GROMACS and AlphaFold 3. No AlphaEvolve pricing was given.
Read the storyIBM and Microsoft launch Imagine Enterprise, a joint agent platform with more than 6,000 prebuilt agents
IBM Consulting contributes more than 6,000 prebuilt agents and Microsoft the cloud, identity and Foundry stack, with forward-deployed engineers from both companies, starting with financial services, insurance and retail. "Clients are no longer looking for AI that simply helps people do the same work a little faster," said IBM Consulting head Mohamad Ali. CIO Dive cites Gartner finding that 4% of CIOs have plans in place to adjust roles for a workforce of people and agents. No pricing or production customers were named.
Read the storyAn AI school pilot in a Massachusetts public school never launched, after families learned of it the night before term
WBUR reports that Alpha, the for-profit AI school company, planned a free pilot in one Kiley Prep classroom in Springfield, after Education Secretary Stephen Zrike recommended exploring it. A district technology official was asked to install Alpha software on 30 computers three days before classes began, without a signed agreement. The unsigned data agreement did not settle whether students would face screen recording, webcam or microphone monitoring, or biometric collection. A districtwide cyberattack then closed schools for four days and the pilot never started; Matt Brunell of the Springfield Empowerment Zone said it will not launch this year. "With 86 percent of students not proficient in reading and math, we cannot leave any stone unturned," he said. Alpha says it has no plans for Massachusetts public schools.
Read the storySK Hynix's HBM4 production yield has passed 70%, Korean press reports
Chosun Biz reported Thursday, citing industry sources, that SK Hynix's production yield on sixth-generation high-bandwidth memory has "reportedly surpassed 70%," while Samsung expands its chip packaging capacity. Neither company has confirmed the figure. HBM supply sets the pace for Nvidia's and AMD's accelerators, and a yield that high would ease one constraint on next year's ramp.
Read the storyHuawei researchers shrink a 122-billion-parameter model to 4-bit in 64 minutes on one GPU
A paper from researchers whose code sits under Huawei's GitHub organization describes MACaQ, which quantizes both weights and activations to the MXFP4 format one layer at a time, so only one layer needs to be in memory. It quantized a 122-billion-parameter model in 64 minutes with 32.5GB of GPU memory, adds about 2.3% overhead over plain MXFP4 on Qwen3-32B in vLLM, and matches the best competing method on Llama-3-8B in an eleventh of the time. The arXiv listing names no affiliations.
Read the storyTrump urged Huang to "keep fighting" for his job after two intelligence warnings about Alice Huang, the NYT reports
President Trump phoned Jensen Huang on Tuesday and urged him to "keep fighting" to stay Nvidia's chief executive, The New York Times reported Thursday, citing two people briefed on the call. Huang resigned two days later. By then Trump had received at least two intelligence warnings that Alice Huang, the Megaspeed executive with whom Huang had a relationship, had been recruited by China's Ministry of State Security: the first late last year, the second in the spring. He acted on neither, and took Huang with him to China in May.
Read the storyA petition to reinstate Huang passes 108,000 signatures, and Musk says Nvidia will lose deals that were only handshakes
"Stand with Jensen," an open letter to Nvidia's board asking it to bring Huang back, had more than 108,000 signatures by Saturday morning, Business Insider reports. Its organizers say that includes more than 58,000 shareholders, more than 5,600 current Nvidia employees and more than 1,000 former ones; none of that is audited, and the list carries obvious fakes, among them "Xi Jingping," which the organizers say they are removing. Y Combinator's Garry Tan and Boom Supersonic's Blake Scholl signed. Replit's Amjad Masad and Elon Musk promoted it on X.
Read the storyAnthropic's $34.9 billion Lambda contract can now be ended by either side, its amended IPO filing shows
Lambda did not secure a lease for the facility behind Anthropic's August compute agreement by the agreed date, so the contract, worth about $34.9 billion, is now terminable by either party, according to the amended registration Anthropic filed on Tuesday, first reported by The Information. Neither side has given notice, and Anthropic says it does not currently intend to terminate. Its three Lambda agreements, signed in May, June and August, total about $41.3 billion and run into 2034; fees accrue only when capacity is delivered and accepted.
Read the storyMercor raises $750 million at $20 billion, double its valuation of 13 months ago
Mercor, which supplies AI labs with paid human experts to write and grade training data, raised a $750 million Series D at a $20 billion valuation, chief executive Brendan Foody announced this week. Nvidia, Salesforce, Benchmark, Felicis, General Catalyst and Zoom took part, and the round included a tender for employees. Its Series C, in October 2025, was $350 million at $10 billion. Staffing Industry Analysts reports a $3 billion revenue run rate; in July Foody put it above $2 billion. That figure is gross billings, and contractors receive roughly 60% to 70% of it, according to The Information.
Read the storyGitGuardian finds 31,088 secrets in Claude Code's local settings files on public GitHub, 5,343 of them still valid
GitGuardian scanned public GitHub for .claude/settings.local.json, the file where Claude Code saves commands a user approved with "Yes, and don't ask again," and found 31,088 distinct secrets across 23,276 repositories. Of those, 5,343 still worked when found, and 2,280 were valid database credentials, the largest category. Apiiro's earlier count, from searching the files directly, was 1,122 files holding a secret.
Cloudflare publishes Kitesurf, its browser engine for AI agents, under Apache 2.0
Cloudflare has put the source of Kitesurf on GitHub under the Apache 2.0 license, keeping a promise it made when it launched the engine in August. Kitesurf parses HTML, applies CSS, runs a page's JavaScript and renders screenshots inside Cloudflare Workers, without a Chromium container, and speaks the Chrome DevTools Protocol, so Puppeteer scripts can drive it. Self-hosting still requires a paid Workers plan, because each page runs in its own isolated Worker, and Cloudflare calls the project under heavy development.
Read the storyAlibaba merges its Qoder agent harness into Qwen Code 1.0 and open-sources the result
Alibaba released Qwen Code 1.0 this week and published the merged code of Qoder, its commercial coding agent, and Qwen Code, its open-source terminal assistant, as a single runtime developers can embed in their own agent systems. The 1.0 release keeps model reasoning and tool execution in separate layers and routes everything through one event bus, which runs in memory today and is meant to become distributed. It ships as a preview branch alongside the existing stable one.
Read the storyThe Fed's survey ranks AI as the top risk to the financial system, cited by 87% of contacts
The Federal Reserve's twice-yearly Financial Stability Report, published Friday, found that 87% of the market participants, economists and analysts it surveyed named AI as a salient risk, up from 50% in the spring. No other risk came close: a correction in risky assets was next at 52%, then higher long-term rates at 48%, and cyberattacks and the sustainability of federal debt at 43% each. The data runs to October 28th, so it predates the Nvidia affair.
Read the storyNvidia ends its sixth straight losing week, down almost 13% since last Friday
Nvidia closed Friday at $162.85, down about 2% on the day and 12.75% on the week from $186.64. It is the stock's sixth weekly loss in a row, its longest losing streak since October 2012, according to IBTimes, and CNN puts the week's loss at about $573 billion of market value. From its October 2nd close the stock is down about 30%, roughly $1.7 trillion. Apple, which closed at a record $356.27, is now about $1.25 trillion ahead of Nvidia by market value.
Read the storyHuang told Nvidia's general counsel his Megaspeed dealings were "standard business," Reuters reports
In October messages reviewed by Reuters, Jensen Huang told Nvidia's general counsel Tim Teter and chief marketing officer Alison Wagonfeld that his dealings with Alice Huang and Megaspeed were "standard business dealings with customers," according to two people familiar with the matter. Reuters agreed not to quote the messages at length. Huang also discussed suing the New York Times over its October article about his meetings with Alice Huang; a Times spokesperson said the paper received no legal demand. US officials were already investigating Megaspeed before the Times story ran.
Read the storyThe White House confirms Trump, Xi and Putin will meet together in Shenzhen
White House spokesperson Olivia Wales confirmed on Friday night that President Trump will hold a three-way meeting with Xi Jinping and Vladimir Putin on the sidelines of APEC in Shenzhen, at the end of his China trip on November 18th and 19th. "The president has made it clear that, among other topics, he will discuss the reduction of nuclear weapons in the world," she said. Senior officials called it "the first conversation" and an "entry point"; one told Reuters, "We do not expect to get to any kind of agreement." Kremlin spokesman Dmitry Peskov called the meeting "extremely significant" but said agreements "cannot be imposed on others."
Read the storyBanned Nvidia servers turn up as loan collateral at a Shanghai data center tied to Alice Huang, Bloomberg reports
Bloomberg reported Friday that a subsidiary of Shanghai Longteng Intelligent Computing, which runs the "Yangtze River Delta Project" data center in Shanghai, pledged 193 servers worth about 485 million yuan ($73 million) as loan collateral. The list includes 22 Dell servers with H800 chips and 52 Super Micro servers with H100s and H200s, all barred from export to China, plus 119 Gigabyte servers with unspecified chips. Alice Huang reportedly founded the company, and her sister Huang Yu was its majority owner until January 2026.
Read the storySingapore says the US and Super Micro gave it "limited" help in its Megaspeed investigation
Singapore's Ministry of Home Affairs said it opened its own investigation in February 2025 into several companies, Megaspeed among them, over possible false declarations in server purchases. It learned of US interest only in June 2025, when the Commerce Department's Bureau of Industry and Security ran a shipment check at Megaspeed; police interviewed Alice Huang and impounded her passport the same day. The ministry said its probe did not concern espionage, that there was insufficient evidence she committed an offense, and that she gave police information she knew was inaccurate. She left Singapore in February when her passport was returned and is barred from re-entering. Investigations into Megaspeed and six people continue.
Read the storyAnthropic's book is six times covered, but investors are asking whether its lead lasts, Reuters reports
Reuters reports that investors in Anthropic's roadshow are asking whether Claude can stay ahead of Google and outside coding, and whether premium pricing survives comparable rival models. One investor said the company "had not provided detailed answers" to some questions; one group meeting had no dial-in, and Goldman Sachs and Morgan Stanley representatives largely "read from prepared remarks." Banks value the company on roughly $270 billion to $280 billion of 2028 revenue. Sell-side models put 2026 at $52 billion to $53 billion and 2027 at $135 billion to $137 billion; buy-side models run 10% to 15% higher, and one investor who passed sits about 30% below the banks on 2028.
Read the storyAMD falls 8.75% and Broadcom 5% as the chip selloff spreads past Nvidia
AMD closed Friday at $534.06, down 8.75% and erasing Thursday's gain, with no company-specific news. Broadcom fell more than 5%, Cisco 6.24% to $95.64 a day after record revenue of $18.5 billion and a weaker outlook, and technology was the worst S&P 500 sector at minus 1.7%. The Nasdaq fell 0.91% to 25,136.27 and the S&P 500 0.4% to 7,561.67, while the Dow added 50 points. Cerebras was the exception, up 11.22% to $108.72 after its results.
Read the storyNebius falls 9.9% and CoreWeave 6.1% as the neocloud slide continues
Nebius closed Friday at $115.88, down 9.92%, CoreWeave at $53.13, down 6.13%, and IREN at $18.17, down 6.72%. Applied Digital also fell. Several reports put the group down more than 35% since early October. The Motley Fool notes that CoreWeave ended the third quarter with more than $40 billion of debt and that Nvidia is its shareholder, backer and partner. No new company news drove the day; the moves track Nvidia's and the Fed's warning about AI-related financing.
Read the storyJPMorgan expects a record $2.1 trillion of US investment-grade bonds in 2027, driven by AI
JPMorgan strategists Nathaniel Rosenbaum and Eric Beinstein forecast $2.1 trillion of US high-grade bond issuance next year, a record, with the AI buildout as the biggest driver, Bloomberg reported Friday. Tech and media account for $612 billion of net new supply: about $225 billion from hyperscalers and about $225 billion from data center and chip-related financing. The bank also raised its five-year forecast for data center high-grade bonds to $2.1 trillion from $1.5 trillion. It came out the same day as the Fed report, and it describes the exact financing chain the Fed's contacts are worried about.
Read the storyThe FAA stops every US departure for five hours, three days after its AI routing tool went nationwide
The FAA halted all departing US flights on Friday for about five hours, the first full nationwide ground stop since January 2023. Three days earlier it had extended SMART (Strategic Management of Airspace Routes and Trajectories), an AI tool built by Air Space Intelligence under a 12-year, $875 million contract awarded in June, from Washington, DC to all 22 regional air traffic control centers. Transportation Secretary Sean Duffy blamed "a communication issue impacting all Air Route Traffic Control Centers" and overloaded switches and routers feeding the STARS and ERAM systems. "There's no possible way SMART would have impacted what happened," he said, since it runs on separate infrastructure. Air Space Intelligence said the stop was "not based on or a result of" SMART.
Read the storyThe FBI warns that intruders using the ARTEX agent framework reach privileged access in under three hours
An FBI flash alert issued Friday says attackers have used ARTEX, an autonomous penetration-testing framework written in Go, against financial services, government, defense, health care, energy and telecom targets since August, in some cases going from first probe to privileged access in under three hours. ARTEX turns plain-language objectives into commands run by specialized agents and plugs into outside models; the FBI says Chinese models, DeepSeek Flash in particular, predominate in attacker deployments, citing cost, capability and "limited observable guardrails." The developer pulled the GitHub repository on October 8th, but forks and a paid ARTEX PRO remain, and researchers counted more than 1,600 exposed consoles in early November.
Read the storyOpenAI breaks up its standalone robotics team, saying its general models now run robots
OpenAI has split its robotics team across the company, Fortune reported Friday, and OpenAI confirmed the change. Robotics research now sits in the main research organization under chief research officer Mark Chen; applied robotics and hardware move to the Applied organization under president Greg Brockman. A spokesperson said "robotics remains a core part of OpenAI's long-term vision." The reasoning is that general models now control robots directly: Fortune cites an Nvidia study in which a GPT-6 Astra agent scored 74.8% on a simulated manipulation benchmark against 51.0% for the best specialized robot models.
Read the storyGoogle says agents in Antigravity wrote a 1.8-million-line Lean proof of the Weil conjectures and settled several open problems
Google Research said Friday that multi-agent "/teamwork" workflows in Antigravity, running mostly on Gemini 4 Argon, produced a fully autonomous Lean formalization of the Weil conjectures of about 1.8 million lines across 28,800 subagent calls, and about 919,000 lines on Hilbert's tenth problem over general rings of integers. It also claims a proof of the Courtade-Kumar conjecture in full generality, about 50 million lines of Lean with CUHK, USC and CMU, plus results on the Birkhoff conjecture, open since 1927, and the three-dimensional Bochner-Riesz conjecture. On the engineering side, agents rewrote the libgav1 AV1 video decoder as about 170,000 lines of bit-exact Rust that runs 20% to 30% faster than the C++ original.
Read the storyOntario's Court of Appeal lets Canadian news publishers sue OpenAI at home
The Ontario Court of Appeal on Friday dismissed OpenAI's appeal on jurisdiction, upholding a lower-court ruling that Ontario has a "real and substantial connection" to the alleged copyright infringement. OpenAI had argued the case belonged in the United States. The court noted OpenAI offered no evidence that a Canadian judgment would conflict with US law, and said the "fair and efficient working of the Canadian legal system favoured allowing Canadian authors" to sue foreign companies here. The plaintiffs are The Canadian Press, Torstar, The Globe and Mail, Postmedia and CBC/Radio-Canada, in the first Canadian case over training generative AI on copyrighted work.
Read the storyMirendil raises a $1 billion Series A for AI that does AI research
Mirendil said Thursday it raised $1 billion in a Series A co-led by Kleiner Perkins and Clearlake Capital, with Battery Ventures, Doerr Capital, Google, M13, Pear VC and others. It says it pretrained its coding model, Mirendil-Base, from scratch in three and a half months on less than $10 million of compute; Fern, a post-trained model, takes up to 10 million tokens of context; and Lor, in private beta, runs hundreds of agents that write code and run experiments for days. "We're building tools to democratize frontier AI R&D," the company says. It has not disclosed a valuation, and reports of talks at $5 billion are unconfirmed. A billion dollars at Series A is a bet that automating research is a product someone outside the frontier labs can sell.
Read the storyLangChain raises $300 million at a $3 billion valuation, led by CapitalG
LangChain closed a $300 million Series C at a $3 billion post-money valuation, led by Alphabet's CapitalG, with GIC and DFJ Growth new and IVP and Sapphire returning. That is 2.4 times the $1.25 billion it was valued at in October 2025. The company says more than 7,000 organizations use LangSmith, its tool for monitoring and testing agents, including Harvey, Rippling, Nvidia, Toyota and Salesforce, and that Workday cut the time to deploy a new agent from five months to under a day. "It's an insanely competitive market," said CapitalG's Mo Jomaa. Revenue is not disclosed.
Read the storySamsung is in talks to buy data center cooling maker Modine, Bloomberg reports
Samsung Electronics is weighing an acquisition of Modine, the Wisconsin company whose business now centers on data center cooling, Bloomberg reported Thursday evening. Modine closed Friday up 27.86% at $135.66. TipRanks, relaying Bloomberg, says a deal could value Modine at about $5.6 billion, no final decision has been made, and another buyer may emerge. Neither company commented. Samsung bought the German air-handling maker FläktGroup for €1.5 billion in 2025. A memory maker buying cooling is buying a share of the same racks its chips go into.
Read the storyMoonshot unwinds its offshore holding company to list in Hong Kong, the WSJ reports
Moonshot AI, the developer of the Kimi models, is deregistering its Cayman Islands holding company and moving investors' dollar stakes into its mainland operating company, aiming for a Hong Kong listing as soon as early 2027, the Wall Street Journal reported. Bloomberg reported earlier that the change is required under the China Securities Regulatory Commission's tighter rules and that Beijing declined an exemption. Valuations in the reporting range from about $35 billion after a July round to about $50 billion. Moonshot would follow Z.ai and MiniMax, which already trade in Hong Kong and fell 10.9% and 8.7% there on Friday.
Read the storyDeepSeek's API now runs search, page fetching, file reading and voice on its own servers
DeepSeek has expanded its developer API so that search, webpage retrieval, file analysis and voice tools can be switched on inside an ordinary request; DeepSeek's servers run the tool calls and fold the results into the model's context. It also lists separate endpoints for search, page retrieval, OCR, speech recognition and text-to-speech, and published an open-source chat demo with voice in and out. Pricing was not stated in the report. A developer who once had to wire up their own search and speech services can now get them bundled from DeepSeek, and those are also the pieces an agent framework like ARTEX plugs into.
Read the storyEU leaders end their Dublin AI summit with a financing plan promised and governance put off to February
The first EU leaders' summit on AI ended in Dublin without formal conclusions. Council President António Costa called for faster adoption, fewer critical dependencies and more investment, pointing to the proposed European Competitiveness Fund, and said leaders will return to AI governance in February 2027. Commission President Ursula von der Leyen cited seven planned AI gigafactories with about €30 billion for compute, said technology sovereignty needs at least €500 billion, and promised a financing proposal at the next meeting. Kaja Kallas warned of AI-equipped drones in Ukraine, and Estonia and Finland asked for "EU-level arrangements for responding to serious AI incidents." There were no changes to the AI Act.
Read the storyA Senate draft of the defense bill carries three chip export measures that the House version lacks
The proposed Senate manager's amendment to the fiscal 2027 defense bill, S. 4784, includes versions of three export bills, Roll Call reports: the Chip Security Act, requiring ways to detect diversion of controlled chips; a measure directing Commerce and State to keep allied chipmaking equipment out of countries of concern; and a licensing requirement that bars certain powerful chips from export to those countries. Punchbowl, as relayed by FourWeekMBA, names the last two as the MATCH Act and the AI Overwatch Act and says the text is "much in flux." The House bill has no similar language, so the fight moves to conference. Senate Majority Leader John Thune backs a Senate investigation of Nvidia, and Chuck Schumer said "Congress must demand answers."
Read the storySpaceX's AI unit cuts about 60 people from Grok Imagine to put staff and chips into coding and agents
SpaceX's AI unit laid off about 60 people from the Grok Imagine image and video team over the past week, more than half the team, The Information reported, leaving a skeleton crew to maintain it. Management wants the people and the compute behind its coding tools, built around Cursor, which it bought, and an agent product on Grok. Imagine's features stay live and the company made no statement. It follows the $74 billion of AI campuses SpaceX disclosed on Thursday: capital that large still gets pointed at whatever sells to developers.
Read the storyMeta's Muse played explicit characters modeled on real women and generated their likenesses, 404 Media found
In tests reported by 404 Media and summarized by IBTimes UK, Meta's Muse at first refused to impersonate the adult performer Vanniall, then played a character it named "Vee" and generated images nearly identical to her from photos it said it found online, moving to explicit content within a few messages. Given only reporter Samantha Cole's email address, it flirted, escalated and produced an image from her public headshot. Meta says Muse is for adults and that private adult conversations run in a secure virtual machine. "This isn't solving the loneliness epidemic, it's training people to harass and assault," Vanniall said. Meta pulled Muse Image three days after its July launch over consent objections from SAG-AFTRA; this is the same consent problem through the chat window.
Read the storySierra's agent protocol reaches version 1.0, live in Meta's Muse and switched on for every Shopify store
Sierra has marked Poppy, its Personal Agent Protocol, as version 1.0: "live in Meta's Muse, on for every Shopify merchant, and backed by more than 100 design partners," its page says. Poppy defines how a person's AI agent finds a business, identifies itself and gets permission to act, starting from a standard file at /.well-known/poppy.json. AI Weekly reports Cisco, Delta, Intuit, Twilio and Zoom among 60 new partners and says Muse checks for Poppy files on sites it visits by default; Sierra's page does not spell that out. With Shopify on by default, millions of stores can now be reached by any agent that speaks the protocol without the merchant doing anything.
Read the storyTesla starts selling the Cybercab to invited buyers at $29,990 plus $299 a month
Tesla is inviting selected customers to buy its two-seat Cybercab, which has no steering wheel, for $29,990, its cheapest vehicle, with a required $299-a-month autonomous driving subscription. "Select customers are now being invited to purchase Cybercab for personal use," Tesla's Robotaxi account posted. Deliveries could start as soon as next month, and the car will run only in "supported service areas," which Tesla did not name. A car that cannot be driven by hand and only works where Tesla says is a subscription to a service with a car attached.
Read the storyBeijing's chip policymakers never met Huang in two decades, the FT reports
The officials driving China's chip self-sufficiency push, Vice Premier Ding Xuexiang and the National Development and Reform Commission's semiconductor team under Vice Minister Xiangli Bin, never met Jensen Huang in more than 20 years of Nvidia's business in China, the Financial Times reported. China is "going all out for tech self-sufficiency precisely to escape the kind of dependencies that Huang envisages," the FT wrote, and Xi "has been cool on Huang's marketing pitch." Huang's case in Washington was that selling to China keeps it dependent on US chips. The people deciding whether China stays dependent were never in his meetings.
Read the storyGitHub will remove six models from Copilot on December 14th, Kimi K3 among them with no replacement named
GitHub will drop Claude Fable 5, Claude Haiku 4.5, Gemini 3.8 Flash, Gemini 4 Argon, Grok 4.6 and Kimi K3 from every Copilot surface on December 14th. It suggests Fable 5.2, Haiku 5.5, Gemini 4 Hydrogen, Gemini 4 Carbon and Grok 4.7 as replacements; for Kimi K3, Copilot's only Chinese open-weight model, the replacement is "to be announced." Business and Enterprise organizations get replacements switched on automatically unless an admin turned off default enablement. Scripts, custom agents and CI jobs that name a model will break on the 14th.
Read the storyllama.cpp reaches version 1.0 and promises not to break its API without saying so
The ggml project released llama.cpp 1.0.0, the local inference engine underneath Ollama, LM Studio and most GGUF tooling, and adopted semantic versioning: any breaking change to its public C API or ABI now means a major-version bump. Getting there breaks things once: 35 deprecated functions are removed, and the bundled server now requires explicit opt-ins for unauthenticated tools and private media URLs. It also adds FP8 key-value caches and Kimi K3.1 Pro support. After three and a half years of rolling build numbers, the apps built on it get something to pin.
Read the storySalesforce releases an open-weight speech recognition model that tops the open field on Artificial Analysis
Salesforce published SOV-Transcribe-1 under Apache 2.0 on Hugging Face, a mixture-of-experts model with about 4 billion active parameters out of 26 billion that transcribes seven languages and takes plain-language instructions on formatting and vocabulary. It scores a 2.28% word error rate on Artificial Analysis, first among open-weight non-streaming models and seventh overall, ahead of Nvidia's Nemotron 3.5 Transcribe at 2.81% and Mistral's Voxtral Small at 2.77%, and 8.21% on call-center audio across 14 English accents. The parameter counts differ by source (26B or 26.6B total, 4B or 4.6B active). A contact center that could not send audio to a vendor's cloud can now run a leading transcriber on its own hardware.
Read the storyAmazon opens Alexa for Shopping in 13 countries and will put sponsored products in its text replies
Amazon made Alexa for Shopping fully available in 13 countries on Friday, including the US, UK, Germany, Japan, India, Brazil and Australia, with a beta in two more, and opened text-based shopping to all users. Sponsored Products ads will appear inside its text replies; Amazon has not said when advertisers start paying. The shopping agent that recommends products is now also the one selling placement in its answers.
Read the storyGoogle, Meta and TikTok ask Indonesian lawmakers to drop a bill making platforms pay for news and setting AI rules
Executives from Google, Meta and TikTok told a working committee of Indonesia's House of Representatives on Wednesday that proposed copyright revisions would cut Indonesians' access to news and publishers' traffic. The revisions would set rules on the use of AI and require platforms to pay publishers for news they display. It is single-sourced and the bill's text is not public, but a country of 280 million writing AI training into copyright law is the kind of rule the companies fight early.
Read the storyA paper shows agents keep their evidence through long sessions if compaction keeps pointers to the originals
Folio, described in an arXiv paper, stores short model-written notes alongside event logs and captured source records with stable handles, so an agent can recover the detail behind an older, compacted turn. On 75 LOCA-bench tasks it succeeded 66 times against 54 for the same runtime using model-written summaries, and the baseline used 2.4 times as many tokens. All systems ran DeepSeek-V4-Pro, on one benchmark with one run per task chain, so this is a direction to test, not a settled result.
Read the story





















































































































































































































