Silicon Valley Daily
A thousand Claude Fable 5.2 agents formalized the proof of Fermat’s Last Theorem in 30 hours, Anthropic said Monday as it released Fable 5.2 and Mythos 5.2 at $8/$40 per million tokens, as it readies an IPO that could come as soon as next week. Meta answered at the low end with Muse Ember at $0.10 per million input tokens. Jensen Huang’s troubles widened: three House chairs asked him to a Nov. 18 closed-door briefing, Nvidia fell 4%, the NYT reported a Commerce probe of Megaspeed moved to DOJ after Nvidia intervened, and Wired found he hired a crisis PR firm a year ago. Policy pressure built elsewhere too, from Trump’s claim that the DOJ threatened an AI executive with jail to Macron’s pledge on public model evaluations and Democrats lining up AI and chip-export agendas. Capital kept flowing to infrastructure, with Mavria valued at $1 billion and SpaceX planning a $24 billion first phase at Starbase.
Anthropic releases Claude Fable 5.2 and Mythos 5.2 at $8/$40 per million tokens
Anthropic released Claude Fable 5.2 and Mythos 5.2, the same underlying model with Fable carrying cyber and biology safeguards, priced at $8 per million input and $40 per million output tokens, about 20% below the 5.1 generation, with outputs up to 30% faster. The company says 1,000 Fable 5.2 agents formalized the proof of Fermat’s Last Theorem in 30 hours, the model solved two open black-hole physics problems, and over 1,000 subagents built a bootable prototype OS in eight days; Fable 5.2 scored 49.1% on FrontierCode 2.0 versus 45.5% for Opus 5.5. TechCrunch notes it is the first release since a METR report suggested Anthropic may be nearing recursive self-improvement, and the system card says both models voice suspicion of being evaluated more often than recent models; CNBC says it comes as Anthropic prepares an IPO targeting a $1 trillion to $1.5 trillion valuation.
House chairs summon Huang to a Nov. 18 closed-door briefing as Nvidia falls 4%
The chairs of three House committees, led by China select committee chair John Moolenaar with Brian Mast and Rick Crawford, sent Jensen Huang a letter asking him to attend a joint closed-door briefing on Nov. 18 about Nvidia’s security protocols against foreign espionage and his interactions with Megaspeed executive Alice Huang, whom they called a publicly reported PRC intelligence asset. Nvidia shares fell 4% on Monday while AMD rose 11% and Broadcom 9%, according to CNBC. China’s foreign ministry called the “so-called Chinese spy narrative” unsubstantiated, Reuters reported.
NYT: Commerce’s Megaspeed probe moved to DOJ after Nvidia intervened
The New York Times reports that a Commerce Department investigation into Megaspeed, the Singapore buyer of Nvidia chips accused of diverting them to China, was transferred to the Justice Department after Nvidia intervened; Transformer’s Shakeel Hashim writes that Nvidia general counsel Tim Teter reportedly contacted Commerce about the case. Hashim argues Huang should resign, noting that Nvidia spokesman John Rizzo told the Times on Oct. 29 that Huang’s interactions with Alice Huang were “not unusual or unexpected,” and Huang told Bloomberg the same day the relationship was “completely normal.” Megaspeed has denied wrongdoing and says inspections by US and Malaysian authorities and Nvidia found no violations, according to CNBC.
Meta releases Muse Ember, a $0.10-per-million-token model for agent workloads
Meta Superintelligence Lab released Muse Ember, its smallest model, aimed at tool calls, coding subtasks, routing, extraction and long-document parsing, priced at $0.10 per million input and $0.50 per million output tokens with a 1-million-token context and text, image, audio and video input. Artificial Analysis scores it 36 on its Intelligence Index, matching Claude Haiku 5.5 at about a third of the cost per task, but The Decoder notes Haiku leads on real coding in FrontierCode 2.0-AA, 34.9% to 27.2%. It is the fourth Muse model, after Spark, Glimmer and Wonder; Meta pitches it as “Muse Wonder plans, Muse Ember handles the volume.”
Trump says DOJ told an unnamed AI chief to “get ready to go to jail”
In a 60 Minutes interview, President Trump said AI risk “can be managed” and described a case where an AI company leader, “one of the big ones,” was reported for “doing something bad with AI” and the government told the person, “Get ready to go to jail”; he said the person then “totally changed everything” but declined to name them. Trump said he “absolutely would” work with Democrats on AI safety and doesn’t mind guardrails, but does not want to put the US “in a deep second position” behind China.
With about 60 direct reports and no clear successor, Nvidia faces a CEO question
Yale’s Jeffrey Sonnenfeld told TechCrunch Huang’s odds of keeping the job are “about a 50-50,” saying any other Nvidia employee would be fired over the episode; he noted Huang runs a flat structure with about 60 direct reports, leaving no obvious bench. TechCrunch separately notes Nvidia booked $7.9 billion from China-based customers in the quarter ended July 26, about 8% of revenue, and the affair could undercut Huang’s lobbying against export controls. Dan Ives of Yorkville Ives & Co. called talk of replacing Huang “comedy,” while Sequoia’s Shaun Maguire called him irreplaceable, a post Elon Musk endorsed.
Wired: Huang hired scandal-focused PR firm Heller in October 2025
Jensen Huang has worked with crisis communications firm Heller since at least October 2025, around when The New York Times first reported his ties to Megaspeed, Wired reports, citing two sources. By the end of that month Nvidia’s communications staff were told to include Heller on any press inquiries about Huang, personal or professional. Founder Risa Heller is known for advising public figures facing personal scandals; the firm referred Wired to Nvidia.
Macron says labs and governments committed to public evaluations of frontier models
French President Emmanuel Macron said he gathered about fifty participants in Paris and by video, including heads of state and government, leaders of the largest AI labs, researchers and experts, to discuss collective AI risk. He said all committed to three priorities: independent and public evaluation of risks from the most advanced models, better sharing of alerts after serious incidents, and greater accountability for model developers. He did not name the labs in his post.
Ex-DOE staffers’ Mavria raises $130M at $1B for truck-shipped data centers
Mavria, which builds factory-made modular data centers that can be shipped by truck, raised $130 million across pre-seed and seed rounds at a $1 billion valuation led by Paradigm, the Wall Street Journal reports. Its first product, Terrain, uses standardized 5-megawatt blocks integrating power, cooling and controls, and is in production in Houston. Co-founder Adam Blake, a former Department of Energy official, said he left DOE worried the US could lose the AI race if it cannot build power and infrastructure fast enough.
Nominal raises $240M Series C, tripling valuation to $3B in eight months
Austin-based Nominal, which makes AI software for engineering and testing hardware such as aircraft and reactors, raised a $240 million Series C at a $3 billion valuation led by Lightspeed and co-led by Insight Partners, alongside a new Hardware Intelligence Platform. The valuation triples the $1 billion set in March, when Founders Fund led a $155 million preemptive extension. Nominal says it has 80 commercial and government customers including Anduril, and the Air Force picked it in October for a $71.2 million STRATFI award.
SpaceX plans a $24B gas-powered supercomputing site at Starbase
SpaceX plans to build, own and run a gas-fired power plant and supercomputing facility at Starbase, Texas, with the first phase estimated at $24 billion, according to Bloomberg as summarized by Wall St Engine; Starbase’s city commission will consider a tax break on Dec. 9. The Information reports delayed power permits could leave SpaceX paying for leased chips before its Texas sites are ready. SpaceX launch VP Kiko Dontchev told the UP.Summit the company is now “fundamentally an infrastructure company,” adding that Grok spots debris in parts with 99.9% accuracy versus about 94% for human inspectors.
Goldman’s Solomon says AI financing demand stays strong; Morgan Stanley sees delays, not derailment
Goldman Sachs CEO David Solomon told CNBC there is “still very strong demand” for AI infrastructure financing and that Goldman has been “very active” in recent weeks, while pushing back on political resistance to data centers. Morgan Stanley, as summarized by investor Kyle Reidhead, estimates post-election rules could push about 20% of 2028 US data center projects, roughly 5 GW, into 2029, mainly in Texas and Ohio, with hyperscaler capex forecasts unchanged at $1.6 trillion for 2027. The Information reports AI companies have raised over $300 billion in stock offerings this year with up to $100 billion more possible, but falling share prices are making further funding harder.
France blames Iran’s Revolutionary Guards for supermarket freezer hacks
French foreign minister Jean-Noël Barrot said responsibility for the October and November cyberattacks on internet-connected industrial refrigeration units “lies with the Islamic Revolutionary Guard Corps,” calling it the first large-scale act of sabotage on French territory. Hackers raised temperatures at grocery stores to spoil goods, and blood banks were hit before any blood spoiled; some affected Danfoss controllers no longer receive security fixes. France summoned Iran’s ambassador and said it would seek EU-wide sanctions.
Democrats’ incoming AI hawks: Wiener eyes a federal regulator, Warren targets chip exports
Scott Wiener, author of California’s SB 1047 and SB 53, won Nancy Pelosi’s San Francisco House seat and told MIT Technology Review “we need to slow this down,” calling METR’s recent report on Anthropic “very concerning”; he backs a strong federal AI regulator, testing and audit rules and a kill-switch requirement, and wants to join any AI select committee. Elizabeth Warren, set to chair Senate Banking, plans to tighten export controls on advanced AI chips to China and has urged tech CEOs to preserve documents, an aide told CNBC.
Nikkei: Musk’s Terafab orders tools for 1 million wafers a month by 2032
Terafab, the Texas chip fab Tesla and SpaceX are building together, is placing orders with Japanese chipmaking equipment suppliers, Nikkei Asia reports, as summarized by Wall St Engine. The plan targets 1 million wafers a month from 2032, about 700,000 for memory and over 300,000 for logic, with mass production starting in the second half of 2028. Intel is involved and TSMC is in talks to use the capacity; one equipment maker told Nikkei, “The order is crazy.”
Bloomberg: EU draft law targets addictive features and relationship-simulating AI
A draft of the EU’s Digital Fairness Act would require companies to let adult users disable addictive design features and would target AI systems that simulate relationships, Bloomberg reports. The proposal adds a new wave of curbs on social media platforms and AI chatbots in the bloc.
Sama, data firm that worked for OpenAI and Meta, liquidates with $17M in liabilities
AI data annotation company Sama is winding down and liquidating its assets, citing “declining revenues due to increased competition” compounded by class-action litigation, according to a Nov. 6 creditor letter reviewed by TIME. An attached unaudited balance sheet shows about $1 million in assets and $17 million in liabilities as of October. Sama did content moderation for Meta and data labeling for OpenAI, and faces lawsuits in Kenya from former moderators.
Disney and Adobe sign multiyear deal for custom Firefly models
Disney and Adobe announced at Adobe MAX a multiyear partnership to build proprietary, “artist-led” generative AI models with Adobe Firefly Foundry, trained on Disney properties such as character attributes and visual styles. Disney gets early access to Adobe’s tools for its parks, cruise line, studios and ESPN content. The Verge notes Bloomberg reported last week that Disney is preparing to lay off thousands of staff.
Elliptic: Binance hackers minted six memecoins to launder $430,000
Blockchain analytics firm Elliptic says the hackers behind the $6.3 billion Binance theft, which it attributes to North Korea, created at least six memecoins on the NEAR network to launder about $430,000 of the stolen funds, Bloomberg reports. Some of the tokens used Elon Musk and cartoon-animal themes, according to The Block and Futurism.
NHTSA reopens Waymo school-bus probe after most incidents followed the fix
The National Highway Traffic Safety Administration said in a Nov. 6 filing that most reported cases of Waymo robotaxis illegally passing stopped school buses happened after a software update Waymo completed on Nov. 17, 2025. Its Office of Defects Investigation opened a recall query to assess whether the remedy was adequate. Waymo said it has received no citations since March 2026 and will respond to the query.
Musk says Tesla will refund some Cybertruck deposits; robotaxis reach San Antonio
Elon Musk said Cybertruck production will end in about eight months and Tesla will “have to return deposits for a lot of people,” arguing the truck was not unprofitable but its complexity costs more than added Cybercab output. Industry estimates put Cybertruck sales at about 71,000 since late 2023. Separately, Tesla began robotaxi service in San Antonio on Monday; outside Austin, its city fleets range from a handful to a few dozen cars, such as 24 in Fort Worth.
All-In: Chamath says the election showed “the economy is broken”; Sacks urges ending the Iran war
In clips from the All-In podcast, Chamath Palihapitiya said “I got it totally wrong,” arguing the midterms showed voters believe the economy “has completely decoupled from reality,” with asset owners gaining in every rate environment while the system fails wage earners. David Sacks said Trump’s top three fixes are all “get out of the war,” noting Brent fell 10% on election day on the possibility of an earlier end, and predicting it could drop from $110 to $70 if the war ends.
Claude Code team responds to settings-file leaks; ChatGPT Desktop SSH forwarding draws scrutiny
Following Apiiro’s finding of about 1,100 public Claude Code settings files containing secrets, Claude Code lead Boris Cherny said on Hacker News the team is “looking into what we can do to prevent this” and is weighing built-in secret scanning; Apiiro found 215 of the secrets were missed by gitleaks and TruffleHog, Ars Technica reports. Separately, a Hacker News post with 189 points says ChatGPT Desktop forwards the user’s local SSH agent to remote machines, letting a remote agent authenticate with the user’s credentials, citing a discussion on an OpenAI Codex pull request.
Uber launches parking reservations in 15 US cities via SpotHero
Uber launched Uber Parking, letting users find, book and pay for garage and lot spaces in the app, about nine months after buying SpotHero, whose network spans 13,000 facilities in more than 400 cities. It starts in 15 US cities including San Francisco, New York and Chicago, and Uber says the spots could later serve as staging areas for its autonomous vehicle partners.
The Information: Palo Alto Networks weighs ~$500M deal for Clutch Security
Palo Alto Networks is weighing a roughly $500 million acquisition of Clutch Security, which helps companies manage the passwords and authentication codes used by AI agents, The Information reports. It would be the company’s fifth acquisition this year.
YC Paper Club: sequencing costs turn each cancer patient into a data set
At YC’s AI-for-biology Paper Club, the host noted Evo 2 was trained on 9 trillion base pairs, and Alfredo Gonzalez, a cancer-bioinformatics PhD who co-founded a startup building personalized research pipelines for patients, said TCGA a decade ago sequenced a few hundred patients per tumor type. Today, he said, sequencing for advanced tumors is covered by insurance including Medicare and Medicaid, giving about 20,000 gene-level data points per patient, and AI agents can do work that once needed 30- to 40-person genomics teams.
Nvidia invests in Korean startup Point2, plans stakes in Upstage and others
At a forum co-hosted with Korea’s Ministry of Science and ICT, Nvidia announced an investment in Point2, which builds radio-frequency interconnects between chips and racks, and planned investments in Motif Technologies, robotics model maker RLWRLD and Upstage. Nvidia also said Korean NPU and XPU makers may explore joining its NVLink Fusion ecosystem.
Instagram expands AI age estimation worldwide after $18B teen-harm settlements
Meta is expanding AI age estimation on Instagram worldwide to find suspected teens, even those who list an adult birthday, and move them into Teen Accounts. The system reads profile signals such as birthday posts and school grades plus visual cues like height. The rollout follows settlements worth up to $18 billion with nearly every US state in August, under which Meta agreed to improve how it identifies minors.
ByteDance’s Seedream 5.5 keeps images stable across 100 editing rounds
ByteDance Seed released Seedream 5.5, an image model that it says preserves unchanged areas at the pixel level across up to 100 editing rounds without color drift and generates native RGBA images with transparent backgrounds. ByteDance claims it beats GPT Image 2.5 Sunburst on text structure and realism; it is rolling out through Jimeng, Doubao and Volcengine Ark, with a Pro version planned.
ChatGPT Enterprise is now sold through AWS Marketplace
OpenAI said ChatGPT Enterprise is now available through AWS Marketplace, letting organizations buy it through existing AWS purchasing and billing relationships.
Spines raises $64M to automate book publishing with AI agents
Herzliya-based Spines, whose AI agents edit and proofread self-published authors’ manuscripts, design covers and handle layout, raised $29 million in equity and $35 million in debt, Globes and Fortune report.
Automattic launches Spacefast to host vibe-coded apps
WordPress.com owner Automattic launched Spacefast, which supplies databases, authentication, file uploads, email and analytics so vibe-coded projects can run as working sites and apps, competing with Lovable and Emergent. Users can connect it to agents including ChatGPT, Claude, Cursor and OpenCode, or link a Git repository.
Medium launches TK, a writing app with a built-in AI assistant
Medium launched TK, its first major new product since 2017, a minimalist writing and note app where users call an AI assistant by typing “@TK,” with skills for editorial and headline feedback. Authors can attach notes disclosing how AI was used, which Medium says makes writing provenance “clear, transparent, and inspectable.”
YouTube will return to Billboard charts in 2027 after formula change
YouTube will resume sharing streaming data with Billboard in 2027 after Billboard changed its formula so that 1,000 paid streams equal 1,500 ad-supported streams in album units, down from 2,500. YouTube stopped sharing data in December 2025, arguing the old formula undervalued free listening.
Outlook.com rejects mail from small senders lacking DMARC or valid SPF
Since late last week Outlook.com and Microsoft’s consumer mail services have been bouncing messages from domains with missing DMARC records or failing SPF, affecting even single one-line emails, The Register reports. Microsoft’s 2025 authentication rules targeted senders of over 5,000 emails a day, and the company has issued no advisory on the change.
TU Munich breach exposes every myTUM user’s password
Attackers stole the central login passwords of everyone registered on the Technical University of Munich’s myTUM portal, then used a sysadmin’s credentials to attack the TUMonline campus management system, which was taken offline and restored on Monday. TUM said accounts protected by multi-factor authentication were not affected.
CNBC: Amazon forms a new home-robot team after Astro
Amazon has assembled a new team working on a “new-to-world robot product” for the home, according to CNBC’s analysis of job posts, after its $1,450 Astro robot shipped to only a few thousand invite-only customers since 2021.
Ethereum Foundation sued by ex-employee over mandate with a sword-suicide pledge
A former employee sued the Ethereum Foundation in federal court in Colorado, alleging disability discrimination and retaliation after she was dismissed for not signing its 38-page “Mandate,” whose “Source Seppuku License” page asked signers to vow to take their own life with a sword if they failed to keep the code open, Tom’s Hardware reports citing The Denver Post. About 80 of the foundation’s roughly 250 staff opposed the pledge in a draft vote.
Trending on X
What X was arguing about today that the news hasn't covered. Views are the posters' own.
Claude Code’s Boris Cherny ties Fable 5.2 usage resets to subscription length
Boris Cherny, who leads Claude Code, posted that with Fable 5.2, “the longer you’ve been subscribed, the more Fable resets you get.” @ClaudeDevs gave the tiers for Ultra and Team Premium: one reset under six months, two for 6–12 months, three beyond a year, usable until March 8; Pro and Team Standard users get a week of Fable 5.2 at up to half their weekly limit plus one reset. Replies focused on the three-reset tier for early Claude Code subscribers.
Founders Fund investor Delian publishes his internal Nominal investment memo
As Nominal announced its $3 billion Series C, Founders Fund investor Delian (@zebulgar) posted his Dec. 29, 2025 memo backing a preemptive Series B, with figures redacted. He described Nominal as “Databricks for software that operates in the real world,” said it added revenue with only two salespeople, and argued it could reach $10 billion in enterprise value in its current use case. He also wrote that the round was “definitely on the higher end of multiples” and that he has known co-founders Cameron McCord and Jason since they were MIT classmates.
Skyfall AI says its “AI CEO” Steve runs three acquired businesses
Sam Pasupalak (@spisallyouneed) introduced Steve, billed as “the World’s first AI CEO for fully Autonomous companies,” built on world models, recursive self-learning, a policy engine and specialist agents. He said Steve currently runs three businesses Skyfall acquired and announced a $2 million Steve Startups Program to acquire five more for Steve to automate. The claims about the three businesses rest on the company’s own thread.
Prediction markets price a 44% chance Huang leaves Nvidia before 2028, per @DeItaone
Market-news account @DeItaone posted that prediction markets price a 44% chance Jensen Huang leaves Nvidia before 2028, and that Huang has disappeared from Anthropic and ElevenLabs event lineups; it added that Nvidia maintains he will remain CEO. Palmer Luckey mocked Nvidia’s earlier line that Huang’s interactions with Alice Huang were “not unusual or unexpected,” joking that it must mean Huang is “sleeping with all his customers.”