Silicon Valley Daily
Intel closed down 28.58% on Friday, its worst day since the 1970s, wiping about $12 billion off the US government's stake and dragging roughly $1 trillion out of chip stocks. The selloff capped a week in which investors started pricing the cost of AI: S&P put Meta's rating on negative outlook, the five biggest spenders' AI commitments passed $3 trillion, and Blackstone now expects Anthropic to be rated junk after its IPO. Reddit fell 29% as Google's AI answers cut its referrals. Investigators say nearly all of Binance's $6.3 billion loss left within 30 minutes, and a tampered Claude Code extension reached up to 5,000 Cursor users. Capital still flows to infrastructure: Baseten is raising about $2 billion at up to $30 billion, Salesforce is buying defense startup Rune, and four AI labs agreed to honor one opt-out signal for publishers.
Intel closes down 28.6%, its worst day since the 1970s
Intel shares closed down 28.58% on Friday after the company guided below expectations, citing a plant shutdown in Israel and a weak PC market, CNBC reports. The US government's 10% stake lost about $12 billion in value, and Nvidia's stake about $6 billion. The PHLX semiconductor index fell more than 9% intraday and roughly $1 trillion in chip market value was erased, Tom's Hardware reports, citing Benzinga estimates; the Nasdaq fell 2.9%.
Binance thieves took nearly all $6.3B in 30 minutes, Elliptic says
Almost all of the $6.3 billion stolen from Binance was taken in the first 30 minutes, Elliptic told the Financial Times, and Chainalysis said the hackers used automated laundering patterns it had not seen before. TRM Labs and Chainalysis attribute the attack to North Korea's TraderTraitor group. Binance estimates about $3 billion has been frozen so far, CNBC reports, but TRM says the hackers quickly sold USDT and borrowed ether, which is harder to freeze.
Tampered Claude Code extension sat in Cursor's marketplace for 11 hours
A maliciously modified copy of Anthropic's Claude Code extension, version 2.1.313, was available in Cursor's extension marketplace for about 11 hours on October 29 and was downloaded by at most 5,000 users, Cursor says. The loader tried to add an attacker SSH key, open a reverse tunnel and send host details to a remote server, BleepingComputer reports. Bugs in its SSH setup left it unable to open access on most Macs and Windows machines, Phorion found. Anthropic says its own systems were not compromised and Cursor has not explained how the package was published.
S&P turns negative on Meta as Big Tech's AI bills reach $3 trillion
S&P Global Ratings moved Meta's AA- rating to a negative outlook, citing elevated capex, negative free cash flow and doubts about its ability to monetize AI; Meta said 'we respectfully disagree', CNBC reports. Alphabet, Meta, Microsoft, Amazon and Oracle have more than $3 trillion in AI spending commitments, including Alphabet's $1 trillion and Meta's $800 billion, Bloomberg reports. Reuters calculates the five returned just 9.3% of cash to shareholders in the first three quarters, a 15-year low.
Reddit falls 29% on weak guidance and Google referral losses
Reddit closed down about 29% on Friday, its worst day since its 2024 IPO, after forecasting fourth-quarter revenue of $1 billion to $1.03 billion versus the $1.05 billion analysts expected, CNBC reports. CEO Steve Huffman said the site lost more than one million Google search referrals a week in October and blamed AI Overviews and AI Mode. Reddit will test requiring users to log in to view content; Huffman said he had been 'holding that line' on logged-out access.
Baseten raising about $2B at up to a $30B valuation
AI inference startup Baseten is raising about $2 billion in two tranches, one at a $26 billion valuation and the other at $30 billion, Bloomberg reports, citing people familiar with the effort. Baseten was valued at $13 billion in June after raising $1.5 billion, so the new round would more than double its valuation in about four months.
OpenAI, Google, Microsoft and Anthropic back one robots.txt signal for AI opt-outs
OpenAI, Google, Microsoft and Anthropic say they will honor a shared 'policy token' in robots.txt that lets publishers opt out of AI training across all participating crawlers at once, The Verge reports. The signal was proposed in the IETF by Kevin Kelley, a standards lead at Anthropic; today sites must set rules for each AI company separately. Meta voiced support but did not say it will honor it. OpenAI says it will implement the signal in the coming months, and on the same day published excerpts of court briefs arguing that fair use supports AI training.
Blackstone now expects Anthropic to get a high junk rating after its IPO
Blackstone, the anchor investor in a $63 billion financing for Anthropic's chips, told investors it now expects the lab to receive a double-B rating, the top tier of junk, after its IPO, TechCrunch reports, citing the FT; Blackstone had earlier said Anthropic would reach investment grade before 2027. A lower rating would make borrowing for compute costlier. CNBC notes the S-1 runs over 300 pages and shows Dario Amodei and his six co-founders each own more than 200 million shares.
Salesforce buys military logistics startup Rune, reportedly for $375M
Salesforce signed a definitive agreement to buy Rune Technologies, which builds AI software that helps the US military predict supply needs, and called it its most significant national-security acquisition, Axios reports. Resilience Media, citing two sources, puts the price at $375 million in cash and stock. Rune had raised $30 million from investors including Andreessen Horowitz and IQT and was last valued at $127 million, per PitchBook. Salesforce says its Missionforce defense unit grew 80% year over year, helped by a 10-year, $5.6 billion Army contract.
Epoch AI: open-weight GLM 5.3 built Firefox exploits for about $1,000 each
Epoch AI researchers say the Chinese open-weight model GLM 5.3 turned public Firefox security fixes into working exploits before the patches reached users, compromising six releases from September 17 to October 26 at about $1,000 in API fees each, The Decoder reports. The chain let a webpage read a visitor's local files. Mozilla separately published a post on defending this 'patch gap' against automated attackers.
Google's first Suncatcher data-center satellite is online, running Gemma 4
Google's Suncatcher prototype satellite, launched on a SpaceX Falcon 9 earlier this month, is online after about three weeks of commissioning, TechCrunch reports. It carries four TPUs and generates up to 1 kW, so it can run one TPU for only about 20 minutes at a time. Its first task was to have Gemma 4 write a haiku about space. The test will measure how the chips handle heat and radiation in orbit.
Toyota will build Waymo robotaxis for Japan and the UK
Toyota will build 2027 bZ4X electric vehicles with Waymo's self-driving system installed at its Takaoka factory in Japan, TechCrunch reports. Waymo will start testing them in Japan next year and plans to use them in other right-hand-drive markets such as the UK, where it has tested Jaguar I-Pace robotaxis in London for months. The deal follows a preliminary agreement signed 18 months ago.
Google in talks with the Pentagon to secure Golden Dome data
Google is in talks with the US Defense Department about a new effort to secure the data that feeds software being built for the Golden Dome missile defense program, Bloomberg reports, citing people familiar with the matter.
Anthropic renames Max 20x as Ultra and lifts its five-hour cap
Anthropic renamed its $100-a-month Max 5x plan to Max and its $200 Max 20x plan to Ultra; prices are unchanged. Verified Ultra accounts no longer face the rolling five-hour usage limit, starting with most existing subscribers, though weekly limits still apply. The New Stack notes the pricing page also lists monthly API credits equal to each plan's price.
OpenAI tightens weapons answers for teens and made 200+ police referrals
OpenAI said it will further limit weapons-related information ChatGPT gives users it predicts are under 18, even when a question looks harmless, Reuters reports. It made more than 200 law-enforcement referrals in the third quarter and will publish a Community Safety Report every six months starting in early 2027. ChatGPT will also give less detail about past mass attacks, including weapons used and how they were obtained, for users of any age.
Apple Intelligence comes to China with iOS 27.3, with Alibaba's Qwen
Apple will launch Apple Intelligence in China with iOS 27.3, it said on its Chinese newsroom; a beta arrives November 4 and the full release in December, The Verge reports. The launch covers earlier features such as notification summaries, live translation, writing tools, visual intelligence and object removal, not the new Siri. Apple says it worked with Alibaba to support its Qwen chatbot in some experiences.
Anthropic researchers build remote root exploit for critical Linux flaw
Anthropic security researchers developed an exploit for CVE-2026-98432, a use-after-free bug in the Linux kernel's IPv4 TCP code that lets unauthenticated attackers run commands as root on servers with Multipath TCP listeners, BleepingComputer reports. The flaw scored 9.8 out of 10 and affects kernels back to version 4.4, including Ubuntu, RHEL and Rocky Linux.
A Claude agent picked a gene therapy and ran surgery on a mouse
An agent running on Claude Sonnet 5 chose a gene therapy from 19 options, picked the brain target and injection coordinates, and operated on an epileptic mouse through a motorized surgical rig, The Decoder reports. Gabriel Ocana-Santero of Exin Therapeutics, who described the work on X, says it is the first time an AI agent has chosen a treatment and delivered it to a living mammal. Mid-surgery the agent detected a calibration error and refused to continue past its safety limits.
Powerlattice raises $86M for chiplets that cut AI chip power
Powerlattice, a Vancouver, Washington startup whose power-delivery chiplet aims to cut AI processors' power needs by more than 50%, raised $86 million, according to an SEC filing reported by GeekWire. It raised $31 million in 2025.
AI law firm Crosby bids for assets of collapsing Weil Gotshal
Crosby, a two-year-old AI-first law firm, has bid for some assets of Weil, Gotshal & Manges, including its trademark, brand and some internal data, and is trying to hire some of its 1,200 lawyers, TechCrunch confirmed. Weil told the New York Times it is preparing to vote on dissolution. Crosby has raised about $85 million, including a $60 million Series B led by Lux and Index. In an a16z video, the founder of legal AI company Eve claims one LA plaintiff firm went from 100 to 500 clients per attorney using its tools.
UK High Court rejects Meta, TikTok and X challenge to Ofcom
London's High Court dismissed a challenge by Meta, TikTok and X to information requests Ofcom made under the Online Safety Act, rejecting their argument that the requests were disproportionate, Reuters reports.
Democrats turn AI into a campaign issue and oppose any bubble bailout
Among 21 potential 2028 Democratic presidential contenders, tech mentions on X nearly quadrupled since January and tech is now the second most common topic in their press releases after jobs and the economy, according to Upswing research for Reset Tech shared with Axios. Separately, Reps. Greg Casar and Maxine Waters introduced a House resolution warning of an AI asset bubble and opposing any bailout of companies or executives hurt if it bursts.
Nvidia pulls the RTX 5080 from its store as Altman says RAM belongs in data centers
Nvidia has removed the RTX 5080 from its online store after pulling the RTX 5090, and retail prices have climbed past 2.2 times its $999 list price, Tom's Hardware reports. Asked about the memory shortage on the Cheeky Pint podcast, Sam Altman said that from a global allocation perspective it is 'quite terrible' for RAM to sit in personal computers rather than data centers, The Verge reports. CNBC ties Intel's weak PC outlook this week to soaring memory and component costs.
Anthropic drops LeetCode interviews; Proton CEO cancels AI re-interviews
Anthropic has removed LeetCode-style coding questions from its interview loop, The Pragmatic Engineer reports. Proton CEO Andy Yen cancelled near company-wide 're-interviews' that would have required staff to explain how they use AI, after a 404 Media investigation into his AI push, according to six employees and all-hands audio obtained by 404 Media. Accenture says AI agents now sift job applicants, with a human making the final decision, The Register reports.
YC asks for prose-first LLMs, vertical models and AI for City Hall
Y Combinator partners set out new requests for startups in videos and on the YC RFS page. Jared Friedman says LLMs have gotten worse at writing as they improved at code and math; Diana Hu wants vertical foundation models that return typed decisions and probabilities. Brad Flora wants AI for data that defense contractors, banks and hospitals cannot send to an API, and Gustaf Alströmer says approving building permits in a tenth of the time would give US cities $16 billion a year in extra tax revenue.
DoorDash says AI agents have placed only 'several hundred' orders
DoorDash CEO Tony Xu said on the earnings call that third-party AI agents have sent 'maybe several hundred orders in total', TechCrunch reports. Customers still order mainly through DoorDash's own app, and Xu said the company is open to working with agent makers.
DOJ charges five in AI-assisted 'ghost student' aid fraud
The US Justice Department charged five people in its first broad crackdown on 'ghost students', fraudsters who use stolen identities to enroll in college and collect federal aid, partly with AI, Ars Technica reports. The DOJ seized more than $750,000 in crypto from a network linked to Kenyan nationals, and the Education Department says it has identified over 830,000 potentially fraudulent aid accounts.
Ransomware operator spent over $7,000 on AI models across seven providers
A ransomware operator used the open-source coding agent OpenCode, configured with 154 model entries across seven providers, to assist intrusions and ran up more than $7,216 in AI usage by October 22, The Register reports. Some of the models had their safety training removed or were marketed as uncensored.
United's CEO doubts Boom's 2029 supersonic target as Boom pivots to turbines
United Airlines CEO Scott Kirby said Boom's first Overture supersonic airliner is 'not going to happen in '29', Ars Technica reports. Boom recently opened a Colorado factory to build natural-gas turbines for AI data centers, and Ars reports that power business has lost a customer.
Analysts see a 150-launch shortfall outside China through 2030
Quilty Space estimates launch supply outside China will fall about 150 launches short of demand from satellite operators and civil space agencies through 2030, Ars Technica reports. Adding US military demand could double the gap if SpaceX does not extend Falcon 9 and Falcon Heavy. Quilty's Caleb Henry said the crisis 'will not go away this decade'.
NASA asks industry how to reboost, service or deorbit Hubble
NASA issued a request for information on Thursday asking companies for ideas to reboost, service or perform a controlled deorbit of the Hubble Space Telescope, Ars Technica reports. Left alone, Hubble will reenter the atmosphere in the early-to-mid 2030s; its orbit was last raised in 2002.
Runware wants to mass-produce AI data-center pods
Runware plans to scale its Romania factory to 30 megawatts of prefabricated data-center capacity a day by 2028, nearly a gigawatt a month, CEO Flaviu Radulescu told TechCrunch. The company aims to bring pods online within 48 hours of installation.
Fujifilm will quintuple magnetic tape output for AI archives
Fujifilm will invest about 10 billion yen, roughly $63 million, to raise its magnetic tape production capacity fivefold, Ars Technica reports. It attributes the demand to generative AI, which is driving growth in archived data.
GCHQ lost a 'very significant number' of drives and still can't count them
Britain's signals intelligence agency GCHQ has lost a 'very significant number' of removable storage drives and still does not know the exact scale, the Investigatory Powers Commissioner's annual report says, The Register reports. GCHQ says it will have automated asset tracking in place by the end of the year and then give a more precise count.
Rivian targets hands-free point-to-point driving in 2027, Uber robotaxis in 2028
Rivian CEO RJ Scaringe spent more of the earnings call on autonomy than on the R2 ramp, The Verge reports. Rivian will give its point-to-point hands-free driving feature to employees this year and customers in 2027, then launch robotaxis with Uber in 2028.
Darkbloom warns Mac owners renting GPUs for AI inference of low payouts
Darkbloom, an Eigen Labs network that lets about 800 Mac owners rent out their GPUs for AI inference, told participants that 'earnings have been low' and may not recover for most providers, The Register reports. The Register attributes the drop to falling token prices.
Satlyt's in-orbit test moves satellites toward shared computing
Satlyt, which builds software to turn satellites into a shared computing network, ran a successful test with the University of Houston on a NASA-funded research project, TechCrunch reports. The software already runs on three satellites in orbit.
Privacy-focused AI assistant Ollie will shut down November 6
Ollie, a family-focused AI assistant that connected to users' email and calendars, will shut down on November 6, its founder and CEO said on X, TechCrunch reports. Ollie had discussed a $7.5 million seed round led by Khosla Ventures in September.