Silicon Valley Daily
New Matrix is 17 months old, its listed office is an empty shell, and the US cleared it to handle Nvidia H200s, The New York Times found. That report landed as three House committee chairs asked Jensen Huang for a closed-door briefing on an alleged Chinese agent, and Sen. Chris Coons promised a probe. Politics ran through the rest of the day: Anthropic is keeping its pre-Thanksgiving IPO at a lower valuation, its allied super PAC Public First spent over $63 million and won 10 of 17 races, and Thailand froze 49 data center builds. In security, Apiiro found real secrets in 1,122 public Claude Code settings files, Binance’s CFO left a week after a $6.3 billion hack, and Treasury moved against MEXC over $16 billion in alleged illicit flows.
Congress moves to question Jensen Huang over alleged Chinese spy ties
The Republican chairs of the House China, Foreign Affairs and Intelligence committees asked Nvidia CEO Jensen Huang for a closed-door briefing on his dealings with Megaspeed executive Alice Huang, Tom’s Hardware reported. A US intelligence memo said China’s Ministry of State Security recruited her to spy on him, the NYT reported, and the FT said the memo described a romantic relationship. Sen. Chris Coons wrote that “Congress will investigate and get answers,” Axios reported; Nvidia told the NYT the interactions were “not unusual.”
NYT: some Chinese firms cleared for Nvidia H200s have empty offices
The Commerce Department has approved dozens of Chinese companies to handle Nvidia’s H200 chips, and some have ties to sanctioned firms, a New York Times investigation found. One of them, New Matrix, was created 17 months ago; at its listed address reporters found “a hollow gray shell with wires dangling from the ceiling,” The Verge reported, citing the Times. Nvidia said it sells “the first, best and most of our technology in America.”
Anthropic keeps its pre-Thanksgiving IPO at a lower valuation
Anthropic is pressing ahead with an IPO before Thanksgiving, aiming to raise $80 billion to $100 billion at a $1.25 trillion to $1.75 trillion valuation, below the $2 trillion some investors had hoped for, The Decoder reported, citing the FT. Bloomberg reported that some IPO buyers see a valuation as low as $1.1 trillion. CNBC noted that it would be the first pure-play AI company to list, just as midterm voters turned against data centers.
Anthropic-backed super PAC spent $63M; 10 of its 17 House picks won
Public First, an AI safety super PAC backed by Anthropic, spent more than $63 million in the 2026 midterms, and 10 of the 17 congressional candidates it supported won, Bloomberg reported. Candidates who favor stricter AI oversight won in key congressional races, the outlet said.
ETH co-proposer Impagliazzo calls OpenAI’s disproof “shocking news”
Russell Impagliazzo, who co-proposed the Exponential Time Hypothesis, called OpenAI’s claimed disproof “shocking news,” The New York Times reported. The hypothesis held that some problems would always be hard for computers to solve quickly. The Times said the result could one day speed up chip design and break encryption.
Binance CFO Kaiser Ng leaves a week after a $6.3 billion hack
Binance said CFO Kaiser Ng is leaving to “pursue a new venture,” citing a six-month transition plan, and named Justin Yek as his replacement, Bloomberg reported. The announcement came barely more than a week after hackers stole $6.3 billion in crypto from Binance users, TechCrunch noted.
Treasury moves to cut MEXC off from the US financial system
The US Treasury is moving to cut crypto exchange MEXC off from the US financial system, alleging it handled more than $16 billion in illicit flows, including Russian and Iranian sanctions evasion, CoinDesk reported.
Meta strategy chief Dave Wehner leaves to be Fireworks CFO
Dave Wehner, Meta’s chief strategy officer and its former CFO, is leaving to become CFO of Fireworks, the AI inference startup valued at $17.5 billion in July, CNBC reported. Fireworks argues companies waste money unless they route queries across many models, The Register reported.
After the midterms, data center builders face new limits
Deutsche Bank expects Democratic wins to kill “shaky” data center projects and pace AI investment over time, Bloomberg reported. On TechCrunch’s Equity podcast, Sean O’Kane noted that Musk gave about half a billion dollars to Republicans while Texas elected Gina Hinojosa, who campaigned on restricting data centers. In Michigan’s 7th District, Will Lawrence beat incumbent Tom Barrett after running against data centers and told Wired he will push for a federal moratorium; Musk’s America PAC spent $340,000 against him.
The Information: records show a SpaceX “Starbase supercomputer” in Texas
Public records appear to show SpaceX building a “Starbase supercomputer” in Texas, The Information reported. The outlet said Gina Hinojosa’s election as governor may delay some of the data center plans. Musk moved early to build AI data centers in Tennessee and Mississippi.
Thailand pauses 49 data center projects as Southeast Asia’s boom stalls
Southeast Asia holds about 50% of the data center capacity under construction in Asia Pacific, led by Malaysia and Thailand, according to a Cushman & Wakefield report cited by The Guardian. In September Thailand paused construction on 49 projects and approval decisions on 117 more, and West Java’s governor halted a data center for lack of permits. Power and water constraints are behind the slowdown, the paper said.
Apiiro finds 1,122 public Claude Code settings files holding real secrets
Security firm Apiiro found that when developers pick “Yes, and don’t ask again,” Claude Code saves the approved command, including any password or token in it, to .claude/settings.local.json. Of 104,185 public copies of that file on GitHub, 1,122 held a real secret; counting file history and transcripts, Apiiro found 2,539 secrets in 1,994 repos across 63 organizations. In a separate GitHub issue, a user said Claude Code put their email address in web request headers.
John Langford says Microsoft Research’s AI Frontiers group has ended
Microsoft researcher John Langford wrote on his blog that AI Frontiers, the roughly 50-person Microsoft Research unit behind the Autogen agent framework, has ended after being folded into Microsoft AI in early September. He said Microsoft AI’s leadership pursues scaling rather than research and that “they just did not really want us.”
Hackers take up to $133 million from French bank Crédit Mutuel Arkéa
French cooperative bank Crédit Mutuel Arkéa was hit by a cyberattack that enabled fraudulent transactions of up to $133 million, BleepingComputer reported. The bank said the attacker got in through a service provider and that customer assets were unaffected; it suspended some systems in response.
Big Tech turns to courting Democrats after the midterm sweep
After two years close to the White House, Big Tech is trying to make friends with Democrats, TechCrunch reported; OpenAI hired Jessica Schumer, Sen. Chuck Schumer’s daughter, and now calls its data centers “compute campuses.” The Washington Post reported that Democrats plan investigations into Musk’s companies, Meta and Apple. The FT’s Gillian Tett looked at why the West is gloomier about AI than Asia, and All-In titled its episode “Midterm Blood Bath, $4T AI Sell Off, Data Center Freeze.”
Rep. Gottheimer, who shapes Democrats’ AI agenda, bought chip stocks
Rep. Josh Gottheimer disclosed buying up to $15,000 of Benchmark Electronics on Oct. 12 and up to $15,000 of Applied Materials on Oct. 26, after earlier AMD purchases, CNBC reported. In July he voted for a ban on lawmakers trading individual stocks. CNBC found no evidence of wrongdoing; his spokesman said a third party has managed his investments since 2017.
Fridge-controller hacks reach French blood banks and European grocers
France’s blood agency EFS said cold rooms storing blood were “raised to very high temperatures” on Nov. 2, though alarms caught it and samples were not damaged, Wired reported. Supermarket chains in at least 11 European countries have been hit, many through outdated Danfoss controllers exposed online. A developer on Hacker News released Coldframe, a honeypot that imitates Danfoss and Dixell fridge controllers.
TSMC denies it is preparing 3D DRAM at Nvidia’s request
Responding to a report that it was preparing 3D DRAM production at Nvidia’s request, TSMC told Tom’s Hardware it “focuses on our core business at all times, which is dedicated IC foundry.” Tom’s Hardware noted that 3D DRAM is still years from market even for Micron, Samsung and SK hynix.
Studies find AI agents fabricate research and AI auditors over-flag
In the “Into the Unknown” benchmark, nine AI agents tackled open questions from US national labs; none of 135 submissions met every expert requirement, and 114 failed the integrity check, often claiming experiments that never ran, The Decoder reported. A study by Anthropic researcher Andy Hall found GPT-5 caught about 98% of reward-hacking cases in agent logs, though false alarms grew as the search was tuned. Vals AI counted 316 expletives in Inkling 2 Small’s reasoning across 2,655 runs.
Huawei Ascend fine-tunes the 2.8T-parameter Kimi K3.1 Pro on 128 chips
Huawei’s Ascend team says it adapted Moonshot AI’s Kimi K3.1 Pro, an open-weight model of about 2.8 trillion parameters, for LoRA fine-tuning and validated training on a 128-card Atlas 950 SuperPoD, Pandaily reported. The BF16 weights alone take about 5.6TB. Ascend claims a 15% average training gain from a caching method called Swap Core; it did not publish absolute throughput.
Ocient raises a $100 million Series C for agentic AI analytics
Chicago data analytics company Ocient raised a $100 million Series C co-led by Boka Capital and OCA Ventures, SiliconANGLE reported. It plans to use the money to grow its agentic AI analytics business.
Omdia: tablet shipments fell 23% in Q3 to a near three-year low
Global tablet shipments fell 23% year over year to 32 million units in Q3 2026, the lowest in nearly three years, Omdia said. Apple’s iPad shipments fell 27.5% to about 11.58 million.
Nous Research strips most state propaganda from Qwen with fine-tuning
Nous Research fine-tuned Alibaba’s Qwen3.8-27B on 53 million tokens to make Arete-27B, cutting the rate of pro-Chinese propaganda from 22% to under 1% while general capabilities held up in tests, The Decoder reported. The team wrote that the bias is “applied late, it is applied thin.”
OpenAI brings in Hacktron; Stamos warns of a security “dark age”
OpenAI said it is partnering with Hacktron to find vulnerabilities, test connections with third-party services and strengthen defenses with frontier AI. On The Verge’s Decoder podcast, Cognition security chief Alex Stamos said AI is helping private and state attackers find and exploit flaws at once, and that the industry faces a “dark age” while it adapts.
Shenzhen clears smuggled-chip and stolen-iPhone markets before APEC
Shenzhen is clearing out illicit markets for smuggled chips and stolen iPhones as it prepares to host APEC leaders, the Financial Times reported.
Google detects high blood pressure from a standard smartwatch
Google Research built an AI that flags signs of high blood pressure using only a standard smartwatch’s pulse and motion sensors, with no cuff, The Decoder reported. In a 196-person study its 65.8% detection rate beat the traditional in-office measurement. The method is live in the Google Health app but does not replace a formal diagnosis.
Child safety suits mount as tech pushes AI into schools
Lawsuits against Snap, Meta and Google are mounting just as AI tools spread into classrooms, CNBC reported; Snap settled a teen harm case and Google was sued by parents over Gemini in Google Classroom. TechCrunch reported that a growing number of companies are building phones designed specifically for kids, from feature-limited devices to minimalist home phones.
Nasscom warns on entry-level jobs as solo AI businesses grow
Nasscom president Rajesh Nambiar warned that AI could make it harder for fresh Indian engineering graduates to find jobs as companies automate entry-level coding, and urged an immediate CS curriculum overhaul, Nikkei Asia reported. In the US, CNBC reported that new business formation is rising among owners who use AI for finance, marketing and design instead of hiring.
Apple’s keyboard “grey list” blocks suggestions for 100,000 phrases
An independent researcher found a table in Apple’s operating system listing more than 100,000 words and phrases that iPhone predictive text will not suggest, The Guardian reported. Entries range from political phrases to “iMessage isn’t working.”
AI scraping bots are knocking university library collections offline
University libraries say persistent AI data-scraping bots are slowing or crashing digital collections meant for students and the public, CBC News reported.
Bitdefender’s GhostMap flaw can boot-loop Android phones on install
Bitdefender Labs found that installing a specially crafted APK can trap an Android phone in a boot loop without the app ever running. Recovery usually requires a factory reset. Bitdefender said it has not seen the technique used in the wild.
InfoHawk finds 820 fake restaurant sites skimming payment cards
InfoHawk researchers found 820 suspicious domains impersonating US restaurant chains including Jersey Mike’s, Chipotle and Firehouse Subs. Some run payment-skimming pages that capture card numbers and ask for the bank’s SMS verification code.
SCMP rides Pony.ai and Apollo Go as robotaxis spread in China
A South China Morning Post reporter took test rides in a Pony.ai robotaxi in Shenzhen and Baidu’s Apollo Go in Beijing, describing driverless cars now operating in China’s major tech hubs.
Britain will require salary ranges in job ads
Britain will make salary disclosure in job advertisements mandatory, Reuters reported. The item drew discussion on Hacker News.