Silicon Valley Daily
Binance says the record $6.34 billion theft began with a fake two-round job interview and a 'coding assignment' its security tools missed. AMD and Super Micro then showed how much money is still flowing into AI hardware: AMD guided next quarter to about $17 billion and Super Micro's revenue rose 298%. The financing strain is visible too. Meta sought up to $25 billion in bonds, Nikkei counts $3.2 trillion of off-balance-sheet AI obligations, and Anthropic's filing shows compute ate 58% of revenue last quarter. Google tied school Search to AI and backed off within hours; Revolut says self-hosted GLM cut its coding costs by 90%; CrowdStrike spent nearly $500 million on agent security, and Apple's 10-K now lists rogue AI agents as a risk.
Binance says a fake job interview and its own AI tool opened the door to the $6.3B heist
Binance published an incident report on Tuesday saying North Korean hackers, whom the FBI identified as TraderTraitor, spent months preparing the October 29 theft of about $6.34 billion. In June a wallet engineer installed a tampered installer; in September another engineer was lured through two rounds of fake job interviews into downloading a 'coding assignment' that carried malware its endpoint tools did not detect. The attackers then used an internal tool to take over the computers of other wallet-team engineers. Co-CEO Richard Teng told CNBC the exchange has replenished 98% of the affected assets.
AMD posts record $13.4B quarter, guides to about $17B as data center nearly doubles
AMD reported third-quarter revenue of $13.38 billion, up 45% from a year earlier and above estimates; data center revenue rose 97% to $8.5 billion. It guided fourth-quarter revenue to about $17 billion versus the $16.2 billion analysts expected, and shares rose about 8% after hours. Lisa Su said she expects revenue to grow more than 80% in 2027, while gaming revenue fell 40% to $773 million and Su said higher memory prices are lifting system costs, so AMD is planning for a down year in consumer markets in 2027.
Google says schools that switch off AI will also lose Search, then softens after backlash
Google said its K-12 products will ship with Gemini, Gemini Notebook and Search turned off by default, and education VP Chris Phillips told The New York Times that schools that do not enable AI will not have access to Search because 'Search and AI are deeply integrated.' Senator Bernie Sanders called it extortion and New York Mayor Zohran Mamdani said Google was using monopoly power to force AI into schools; Mamdani separately told Wired he supports an AI ban for some students and a one-year data center pause. Later on Tuesday Google said on X that schools can turn each of the three products on individually, and that AI Mode and AI Overviews for schools will guide students step by step instead of handing over answers.
Meta seeks up to $25B in bonds as Big Tech's off-balance-sheet AI obligations near $3.2T
Meta sought to raise $20 billion to $25 billion in a bond sale on Tuesday, two-thirds in dollars and one-third in euros, CNBC reported, citing people familiar. The sale follows S&P's move last week to a negative outlook on Meta's AA- rating, a $95 billion equity-raising plan, and a quarter of negative $2.12 billion free cash flow. Nikkei Asia separately calculated that US tech giants' off-balance-sheet AI obligations nearly doubled in six months to about $3.2 trillion.
Fal raises $1.3B led by Bessemer, doubling its valuation to $16B since April
Fal, a San Francisco startup that hosts and runs AI video, image and audio models, raised a $1.3 billion Series F led by Bessemer Venture Partners, with Nvidia, Sound Ventures and T. Rowe Price participating. The round values it at $16 billion, up from $8 billion in April, according to Bloomberg; Fal said its Series E and F together brought in $1.65 billion. Bloomberg reported Fal passed $800 million in annualized revenue earlier this year.
Anthropic's filing shows compute ate 58% of revenue in Q3, up from 48%
Anthropic's updated S-1 shows compute costs reached $10 billion in the third quarter, equal to 58% of its $17 billion revenue, up from 48% in the prior quarter, TechCrunch reported. Revenue per dollar of compute fell to 1.72x from just over 2x. The filing says $5 billion of that compute came under short-term commitments that cost significantly more than its multi-year deals, as the company scrambled to keep up with demand.
Super Micro's revenue nearly quadruples to $19.95B on AI servers
Super Micro reported fiscal first-quarter revenue of $19.95 billion, up 298% from a year earlier and well above the $16.39 billion analysts expected, according to Reuters. The server maker forecast second-quarter revenue above Wall Street estimates on demand for AI-optimized systems, and its shares rose more than 7% after hours.
CrowdStrike buys Fiddler AI and Dreadnode, launches a kill switch for rogue agents
CrowdStrike agreed to buy Fiddler AI, which monitors AI agents inside enterprises, for about $400 million and Dreadnode, which automates security testing of AI models and agents, for about $95 million, The Wall Street Journal reported. It also launched an AI kill switch inside its Falcon Guardian platform that lets security teams stop internal agents, block them from relaunching and cut their network access; a human still has to make the call. The company announced an integration with Anthropic's Claude Cowork as well.
OpenAI hires Uber's Jill Hazelbaker as chief corporate affairs officer
OpenAI appointed Jill Hazelbaker, Uber's president and chief corporate affairs officer, to the new role of chief corporate affairs officer, where she will lead global communications and corporate positioning. Axios reported she will report to Greg Brockman. The hire comes a day after OpenAI named former Army Secretary Dan Driscoll to a national security post.
Revolut cut AI coding costs about 90% by self-hosting GLM instead of Claude Code
Revolut says it cut AI coding inference costs by roughly 90% in under a year by making self-hosted GLM models its default instead of commercial tools such as Claude Code, The Register reported. Ops lead David Abitbol said the fintech now serves GLM-5.3 on 56 Nvidia B200 GPUs for about 2,000 engineers, and that self-hosting cost about a tenth of Codex or Claude Code. He said engineers had been spending $1,000 to $10,000 a month on tokens before the switch.
Canva restructures, affecting 1,271 workers or 22% of staff
Canva CEO Melanie Perkins announced a restructuring affecting 1,271 people, or 22% of staff, merging the product, growth and platform teams into one product organization. Some employees will change roles or be asked to relocate to central hubs, with Sydney becoming home to more product teams; others will leave with at least six months of salary. The Age reported the cuts follow slowing growth, and the move comes a day after Squarespace cut 19% of its staff.
Anthropic open-sources FlashProve to prove which model produced an answer
Anthropic released FlashProve, an open-source prototype that produces a cryptographic proof that a response came from a specific model, so a router or provider cannot quietly swap in a cheaper or stronger one. Labs would publish a fingerprint of each model that does not reveal its weights, and verification is designed to run on a local device. The researchers released a specification and code but said there are no plans yet to deploy it at Anthropic or any other lab.
Traversal raises $120M at a $1B valuation for AI site-reliability agents
Traversal, which builds AI agents that help companies troubleshoot site outages, raised $120 million co-led by Kleiner Perkins, Redpoint and Khosla Ventures at a $1 billion valuation, Forbes reported. Its customers include American Express and Pepsi.
Palantir slips as US commercial contract value grows just 11%
Palantir shares fell 1.65% on Tuesday after its third-quarter report, even though revenue rose 96% to $2.31 billion and US commercial revenue of $984 million topped government revenue for the first time. Analysts focused on total contract value: Palantir closed $1.5 billion of US commercial TCV, up 11%, after a record $2.1 billion in the second quarter that was up 153%. Goldman Sachs said the company attributed part of the slowdown to shorter contract durations.
OpenAI says GPT-6 Astra designed a Codex hardware device in under a week
OpenAI says its hardware team used GPT-6 Astra to design a device for talking to the Codex coding agent in under a week, covering the enclosure, PCB layout, component selection, firmware and interface. A contract manufacturer built prototypes in under a month, but several revision rounds were needed because the model did not always account for manufacturing tolerances, according to The Decoder. OpenAI said an unreleased internal model scores 0.875 on the BenchCAD 2.0 benchmark versus 0.594 for Astra, and that its hardware teams' token usage has grown 23x since February.
Claude Code and Cowork product lead Cat Wu leaves Anthropic
Cat Wu, the product lead for Claude Code and Cowork, has left Anthropic, Business Insider reported. She joined in 2024 and helped scale Claude Code. Her exit follows Claude Code engineer Thariq Shihipar's departure announcement last month.
The Information: OpenAI stayed quiet on China chip exports to avoid angering Nvidia
The Information reported on how Nvidia uses its position as the dominant chip supplier to influence customers and US chip policy. According to its sources, OpenAI took no public stance on chip exports to China to avoid angering Nvidia. The report says the issue surfaced internally last month when OpenAI senior research scientist Noam Brown shared a news story in a company Slack channel.
Shopify drops 8.7% on a GMV miss; Block raises its profit outlook
Shopify shares fell 8.74% after third-quarter revenue of $3.74 billion came in just under estimates and gross merchandise volume of $119.15 billion missed the $120.03 billion expected; subscription revenue and gross profit also fell short. Block, by contrast, reported revenue up 10% to $6.7 billion, Cash App gross profit up 27% to $2.07 billion, and raised its full-year gross profit outlook, sending shares up more than 5% after hours, Reuters reported.
Apple's annual filing adds a risk factor for AI agents acting 'unpredictably'
Apple's annual 10-K, filed Tuesday, added risk language warning that AI agents can act autonomously, including accessing systems they are not authorized to use, executing or altering code, sending communications and completing transactions. Apple wrote that agents 'may behave in unintended or unpredictable ways,' which could expose it to lawsuits or reputational harm. Last month Apple changed full disk access on Macs to reduce risks from installed agents.
Developer says Postman's desktop app installed plugins into Claude Code and Codex unasked
A developer reported on GitHub that opening Postman desktop 12.33.4 on macOS installed Postman plugins into both Claude Code and OpenAI Codex, edited Codex's config file and approved a session-start hook that tells Codex to load Postman's skill for API tasks, The New Stack reported. Postman's documentation says its installer can approve that hook on the user's behalf. The New Stack noted that the GitHub report documents configuration changes, not evidence that Postman collected API data.
Midterms close with $187M of AI-made ads and a fight over Meta suspending No Kings
More than $187 million was spent on nearly 300 political ads generated or enhanced by AI this cycle, the Wesleyan Media Project found, with over 90% of that from Republicans, CNBC reported; there is no federal law requiring disclosure. Platformer reported that frontier labs set up election war rooms. Separately, Meta apologized for suspending the No Kings coalition's Instagram account for four hours during an early-voting push; Meta's Andy Stone said the account had used a Claude tool in a way that made it look like a bot, and a coalition of 14 attorneys general has asked Meta to explain.
Data centers turn to converted jet engines; Drax proposes a £20B site at its biomass plant
Some AI data centers stuck in multi-year grid queues are generating power with retired jet engines rebuilt to burn natural gas, The New York Times reported; xAI is among the users. Rystad analyst Audun Martinsen said such units can get a data center running in 24 months, while a 100 MW grid hookup in Northern Virginia takes about seven years, per the IEA. In the UK, Drax unveiled plans with Brookfield and its Radiant unit for a £20 billion data center of more than a gigawatt at its North Yorkshire biomass station, though it still needs planning permission and a customer.
Pimco's CEO warns AI will hit a downturn; All-In's Friedberg argues every layer is profitable
Pimco CEO Emmanuel Roman told a Hong Kong investment summit that the AI market will see a downturn 'and then things are going to break,' and he has estimated the buildout could create $2 trillion in financing needs over five years, The Verge reported, citing Bloomberg. The Verge noted that debt demand remains strong, with Meta drawing $58 billion of orders for its bond and Microsoft selling $25 billion of bonds. On the All-In podcast, David Friedberg made the opposite case in a segment titled 'Everyone is Making Money in AI.'
Apple will require 18+ ratings for dating apps and romantic AI companions
Apple told developers that all apps offering dating or romantic-relationship features, including AI companions with sexual or romantic content, must carry an 18+ rating. From November 16, developers must add new content descriptors, including whether an app shares a user's real-world location with others. TechCrunch noted that top AI companion apps had been downloaded 220 million times as of last summer.
UK's NCSC expects AI-powered ransomware on critical infrastructure by 2028; mSecure and GitBook report incidents
The UK's National Cyber Security Centre said AI-powered malware is 'highly likely' to feature in a ransomware attack on British critical infrastructure by 2028. Password manager mSecure told customers to change passwords, reset two-factor authentication and possibly remove passkeys after 'suspicious activity,' without saying what was exposed. A GitBook customer posted a support reply saying an attacker 'using AI agents' gained access to some infrastructure credentials on November 2, and Have I Been Pwned put last year's Harvard breach at more than 850,000 people.
Steam reviews now take up to 12 business days as submissions surge
Valve said a 'significant increase' in store and build submissions has pushed review times to up to 12 business days, from the usual three to five, after indie developers complained of missed launch dates. Much of the delay comes from re-reviews, so Valve now requires developers to run games through Steam for at least five minutes and confirm achievements and redistributables before submitting.
Stanford study finds 44% of widely cited LLM findings have already expired
Stanford researchers Owen Queen and James Zou retested 50 widely cited claims about large language models on 20 models released between July 2024 and September 2026, and found 44% have expired and another 34% have weakened, The Decoder reported. Only 16% still hold. Claims about what models cannot do, such as counting letters in 'strawberry' or losing information in the middle of long contexts, expired fastest.
Ubisoft and others bid for Microsoft's Arkane Studios, Bloomberg says
Ubisoft and at least two other gaming companies have offered to acquire Arkane Studios, the French maker of Dishonored and Marvel's Blade, from Microsoft, Bloomberg reported, citing sources. Microsoft's gaming unit is exploring options for the studio before a December deadline.
Ballistic Ventures files to raise $490M across two cybersecurity funds
Ballistic Ventures, the cybersecurity VC firm co-founded by former Kleiner Perkins partner Ted Schlein, filed with the SEC to raise $360 million for its third flagship fund, matching its 2024 fund, and $130 million for a separate vehicle called Fund B, TechCrunch reported. Neither fund had recorded sales at filing.
Baidu builds Apollo World simulator to train robotaxis on rare hazards
Baidu has developed an AI-driven simulator called Apollo World that builds complete virtual journeys from scratch, so its Apollo robotaxis can practice rare, high-stakes hazards such as cyclists running red lights or animals darting into traffic, the South China Morning Post reported. Executives said it covers scenarios that are hard to capture safely in real-world testing.
Microsoft tunes Windows 11 for 8GB PCs as memory prices climb
Microsoft said it is 'raising the bar for 8 GB Windows 11 PCs,' claiming such machines will be up to 20% more responsive and 40% faster at multitasking after memory-management changes. The Register noted the push comes as AI data center demand drives up RAM prices, while Microsoft's Copilot+ PCs require at least 16GB.
Reddit tests ads in search results days after warning of falling search traffic
Reddit began testing ad placements in its search results with a select group of advertisers, days after warning investors about declining search traffic and a weaker revenue forecast. Ads will appear based on factors including relevance to the query; Reddit gave the example of espresso machine ads for a search on 'espresso station.'
California takes $1.42B in broadband funds but rejects the condition to drop net neutrality
California is accepting $1.42 billion from the federal BEAD broadband program to reach 270,571 locations, but told the Trump administration it will not agree to stop enforcing its net neutrality law and other telecom rules, Ars Technica reported. NTIA requires states taking BEAD money to forgo rate regulation and net neutrality rules for funded ISPs, and California reserved the right to challenge that condition.
Netflix adds an 'AI Discover' tab with mood-based recommendations
Netflix is rolling out an AI Discover tab whose first experiment recommends titles by mood, viewing preferences and available time, using curated themes such as 'something light' and 'scare me' with clips from shows and films. Netflix said it is powered by a new intelligence layer that responds to what users tell it in real time.
Nexus Mods makes AI disclosure mandatory as AI addons flood World of Warcraft
Nexus Mods, the largest PC game modding site, will require authors of new uploads to declare whether AI was used in their work, owner Victor Folmann said. Ahead of Wednesday's launch of World of Warcraft: Forever, 404 Media reported that game subreddits are flooded with AI-generated addons copying developers' work, and that Warcraft Logs is changing its API in response.
SETI hands telescope control and signal analysis to AI agents
The SETI Institute and Breakthrough Listen now use AI agents at the Allen Telescope Array in Northern California to steer the telescope and analyze radio signals in real time on Nvidia's Holoscan platform, The Decoder reported. The team says work that took weeks now takes hours, and that it detected a Crab Nebula pulsar in real time with AI for the first time.
Orbital Matter 3D-prints a rigid beam on a free-flying satellite
German startup Orbital Matter said its Replicator 2 satellite 3D-printed a half-meter rigid beam from UV-cured resin in orbit, which executives called a first for a free-flying spacecraft. The satellite launched on a SpaceX rideshare this summer with four printers aboard; the company pitches in-space construction as a way to build larger structures, such as those orbital data centers would need.