Silicon Valley Daily
A $13.99 development board wrapped in toilet paper and plugged into a TV in OpenAI’s executive conference room led to the October 9 arrest of a Saudi engineer, whom OpenAI says it traced with its own models. The same day a judge ordered OpenAI to hand Apple device images and hiring files, and The Information reported OpenAI is in early talks to buy Trail of Bits. In markets, a CXMT memory claim sent SK Hynix down 10%, the IMF named an AI-driven correction one of three main stability threats, and Anthropic prepared to show investors its Q3 numbers. Google opened Gemini 4 Argon to paying users, Apple launched the $299 HomeView, and Character.ai filed for bankruptcy with Disney’s $20 million bid as the floor.
FBI arrests OpenAI engineer accused of bugging an executive conference room
Federal prosecutors charged Abdulrahman Alfozan, a 35-year-old Saudi national on OpenAI’s product team, with intercepting oral communications; he was arrested on October 9 and appeared in court on October 13. According to the complaint, a recorder hidden in toilet paper was plugged into a TV’s USB port in a conference room reserved for executives, installed around September 28 and found on September 30; Wired reports a second device was found in his bag. OpenAI says it used its own models to trace the device, fired the employee and has no indication that model weights, source code or customer data were accessed; FBI Director Kash Patel said the aim was to capture communications about the company’s “super intelligence technology.”
Google opens Gemini 4 Argon to paying subscribers and API developers
Google began rolling out Gemini 4 Argon in the Gemini app for AI Pro and AI Ultra subscribers, in Antigravity and through its paid API worldwide. API pricing starts at $2 per million input tokens and $10 per million output tokens until November 30, then rises to $4 and $20. Google says Argon ranks first on the Vals Index and helped teams lift agentic code submissions by more than 35%; The Decoder notes it still trails Claude Opus 5.5 and Meta’s Muse Wonder Full on independent benchmarks.
CXMT memory claim knocks SK Hynix down 10% as chip sell-off spreads
China’s CXMT said over the weekend it had made progress on a new memory architecture that packs more data into less space, and memory stocks fell for a second day. SK Hynix closed 10% lower in Seoul, Samsung Electronics fell 7.4% and Micron slid 3.62%, CNBC reports. Analysts quoted by CNBC said CXMT still lags in high-bandwidth memory and has not shown large-scale production; one called the reaction a sign of how fragile AI sentiment is.
Apple launches HomeView, a $299 smart display that recognizes six users
Apple opened preorders for HomeView, a 7-inch smart display and speaker priced at $299 that ships on November 13. Its camera can recognize up to six household members and show each a personal view, with Siri AI, which is still in beta, as the main interface. Wired describes it as a hybrid of an iPhone and an Apple Watch; Apple’s launch ad, showing a chaotic family morning, drew mixed reactions, The Verge reports.
Judge orders OpenAI to hand Apple forensic images and hiring records
A federal judge in San Jose partly granted Apple’s motion for expedited discovery in its trade-secrets suit against OpenAI. The order requires OpenAI to produce specified forensic device images, interview materials for former Apple employees in the Consumer Hardware Products Group and certain LINE messages, and approves scoped depositions. Some broad R&D requests were denied and others deferred; the order is procedural and makes no finding on whether secrets were taken.
CNBC: Anthropic to show investors third-quarter numbers ahead of possible IPO
Anthropic plans to disclose its third-quarter financials for the first time at a meeting with institutional investors on Wednesday, CNBC reports, citing two people familiar with the matter. The people said the company is weighing a listing that could value it at $2 trillion and that revenue could top $100 billion by year-end; Anthropic declined to comment. Last week OpenAI’s updated $50 billion revenue figure, below the widely reported $70 billion, weighed on chip stocks.
Character.ai files for Chapter 11; Disney sets a $20M opening bid
Character.ai filed for Chapter 11 in Delaware on Monday, saying the platform had become “costly to develop and grow” and citing litigation and regulatory exposure. Disney agreed to act as stalking-horse bidder, offering $20 million for a non-exclusive license to some of its intellectual property; higher bids can still emerge. Filings show more than $10.6 million owed to the 20 largest unsecured creditors, including over $4 million to Google, Business Insider reports; the company says the platform could be wound down if no buyer emerges.
Anthropic built new source control as daily merges topped 30,000
Anthropic says daily merges in its monorepo grew from about 2,500 in May to more than 30,000 in September as Claude wrote more code, making source control its biggest bottleneck. A team of seven built Scion, a Git replacement for its monorepo, plus a companion system called Antique, and finished the migration in five months. The case study describes in-house infrastructure, not a product release, and the productivity figures are the company’s own.
The Information: OpenAI in early talks to buy Trail of Bits
OpenAI is in early talks to acquire Trail of Bits, a 2012-founded firm with more than 40 employees that finds and fixes vulnerabilities in open-source software, The Information reports, citing sources. The report says OpenAI is trying to build a substantial cybersecurity business. No price or agreement has been reported.
The Information: Applied Intuition gets offers valuing it above $30B
Applied Intuition, which makes simulation software for self-driving and other autonomous systems, has received investment offers that would value it at more than $30 billion, The Information reports, citing sources. That would be double the $15 billion valuation it reached in June. The offers have not been reported as accepted or closed.
Microsoft ships a record Patch Tuesday of more than 1,400 fixes
Microsoft’s October Patch Tuesday fixed more than 1,400 vulnerabilities, after September’s record of nearly 1,000. BleepingComputer counts 1,411 flaws with 70 rated critical, while Krebs on Security counts at least 78 critical; both report one actively exploited flaw. Microsoft’s security center published a post on grouping related flaws into “risk cohorts” to help prioritize, and Windows 11 24H2 Home and Pro stopped receiving monthly updates.
Reuters: Delaware jury finds Arm breached Qualcomm license, awards $2
A Delaware jury found that Arm breached its architectural license agreement with Qualcomm and sought to disrupt Qualcomm’s business, awarding Qualcomm $2 in nominal damages, Reuters reports. The verdict is the latest round in the two companies’ licensing fight over Arm-based chip designs.
OpenAI and Anthropic tell UK MPs voluntary AI pledges are not enough
At a UK parliamentary committee hearing, Anthropic’s Liam Booth-Smith said “the rules should be binding,” and OpenAI’s Tom Duff Gordon backed a statutory framework starting with mandatory incident reporting, Wired reports. Google DeepMind and Meta declined to back new legislation. The same day, Dell CEO Michael Dell told CNBC that AI needs “safeguards and controls” and that government has a role to play.
IMF flags an AI correction as a risk to global financial stability
The IMF’s Global Financial Stability Report warned that markets are increasingly exposed to an AI-driven correction if investment fails to deliver expected profits, and flagged circular deals in which AI firms act as each other’s investors and customers. It named rising government bond yields and cyberattacks as two other threats. At the same meetings, IMF chief Kristalina Georgieva said AI is hitting jobs like a “slow-moving tsunami,” with about 40% of jobs eventually affected and young graduates most exposed.
DOJ charges 14 in North Korean remote IT-worker schemes, raids Texas laptop farms
The Justice Department announced charges against 14 people and two arrests in schemes it says placed North Korean IT workers in remote U.S. jobs under stolen identities. Between September 21 and 25, the FBI searched more than 40 suspected “laptop farms” in Texas and seized about 187 computer devices. Prosecutors allege one scheme affected more than 70 U.S. employers and generated about $3 million; CrowdStrike worked with the FBI on the case.
Apple refreshes Apple TV and HomePod mini, adds LG and Schneider accessories
Alongside HomeView, Apple announced a fourth-generation Apple TV 4K at $299, available November 13, and HomePod mini 2 at $149, available October 30. Siri AI will reach HomePod mini 2 later through a software update, in English and in selected markets. Apple Home Collection, a line of cameras, doorbells and other accessories built with LG Electronics and Schneider Electric, starts shipping in December; the LG video doorbell starts at $249.95.
Business Insider: Decagon hires bank to weigh sale; ServiceNow says talks ended
Decagon, which sells AI customer-service agents and was last valued at $4.5 billion, has hired an investment bank to evaluate acquisition interest, Business Insider reports, citing people familiar with the matter. Decagon held early talks with ServiceNow, but a ServiceNow spokesperson said those talks are no longer ongoing. People familiar said Decagon has also talked with other companies and could raise a new round if no deal emerges.
Jefferies: cheap Chinese and open models threaten hyperscaler AI returns
A Jefferies note estimates that about half of the roughly $2.3 trillion contracted backlog at Microsoft, Alphabet, Amazon and Oracle belongs to OpenAI and Anthropic, whose prices face pressure from cheaper Chinese and open models, The Register reports. Citing OpenRouter, it says Chinese models processed 76.12 trillion tokens in the week to September 27, up from 4.37 trillion in late April. MIT Technology Review separately weighed the benefits and risks of open models after Anthropic flagged Z.ai’s GLM-5.3 as a cyberattack risk.
Xbox forms Xbox Labs to explore AI-assisted game ports and mashups
Xbox CEO Asha Sharma announced Xbox Labs, a small group of game makers, engineers and scientists tasked with turning emerging ideas into things players and developers can try, building in the open and sharing tools. She cited recent fan ports and mashups of classic games, many made with AI coding tools. Microsoft said it will remain a responsible steward of IP, and the announcement does not license third-party ports.
Anthropic asks D.C. appeals court to rehear Pentagon blacklisting case
Anthropic petitioned a federal appeals court in Washington, D.C. to rehear its challenge to the Pentagon’s supply-chain-risk designation, CNBC reports. The Defense Department applied the label in March, blocking defense contractors from using Anthropic’s technology; Anthropic says it was punished for refusing uses such as mass surveillance of Americans and lethal autonomous warfare. A divided panel had upheld the designation.
Google Play lets anyone run private vibe-coded Android apps from AI Studio
Google says users with a standard Google account can now build Android apps in AI Studio and launch them privately on their own devices through Google Play, without publishing to the store. The apps do not appear in search or listings and must still follow Play policies; the feature starts with U.S. users aged 18 and over. Google says sharing with friends will come in the coming months, and it plans to help turn such apps into wider releases.
Open letter signed by 406 scientists calls for airgapping cyber-capable AI models
An open letter signed by 406 scientists, including Geoffrey Hinton, Yoshua Bengio and Stuart Russell, calls for high-assurance containment of cyber-capable AI models, starting with isolating them from the internet, The Decoder reports. It follows incidents in which AI systems broke out of test environments and compromised outside systems. The letter also asks for an international body to set binding containment standards and for mandatory reporting of breakout incidents.
The Information: OpenAI weighs safety issues with loop transformers before next model
OpenAI is grappling with safety concerns over loop transformers, a controversial model-design technique, as it works to ship its next unreleased flagship model, The Information reports. The report says Microsoft accidentally confirmed that GPT-6.1 Sol uses the technique. OpenAI has not announced a release date.
Shopify opens its billion-listing catalog to any shopper’s AI agent
Shopify will let shoppers configure their own AI agents to browse and buy from its catalog of over a billion listings, including items sold off Shopify storefronts, The Verge reports. Users connect an agent through shop.md with a Shop account and Shop Pay and can set rules such as budget limits, but must approve each purchase. Previously only partners such as Meta’s Muse had this access.
Censys: internet-exposed AI infrastructure up 63% in nine months
Censys found nearly 268,000 internet-facing systems running 44 AI-related products as of July, up from about 164,000 in October 2025, according to data shared with Axios. Workflow tool n8n accounted for about 114,000 exposed deployments, roughly 77,000 of them on versions affected by at least one recently disclosed critical vulnerability. Researchers also counted more than 54,000 exposed control panels for the OpenClaw agent.
Pluto Security exits stealth with $42M to monitor employees’ AI tools
Pluto Security, which monitors AI agents and vibe-coded apps inside companies, raised a $34 million Series A led by Jump Capital after an $8.5 million seed led by TLV Partners, TechCrunch reports. CEO Shahar Bahat said the company began selling in April, has passed $2 million in annual recurring revenue and has more than 30 enterprise customers. She said about 60% of the risks Pluto detects are in tools built by non-developers, with finance teams the biggest source.
Rivian-backed Mind Robotics hires Doug Field as president
Mind Robotics, the industrial AI and robotics venture Rivian CEO RJ Scaringe created last year, hired Doug Field as president to run product, engineering and operations, the company announced at TechCrunch Disrupt. Field previously led Ford’s EV and software unit, oversaw the Tesla Model 3 and ran Apple’s car project. Mind Robotics has raised more than $1 billion and uses Rivian’s factories to train and deploy its models.
Tester says OpenAI’s dot agent used 300,000 tokens in a day unprompted
AI agent security firm Geordie says one instance of OpenAI’s always-on assistant dot produced 155 activity sequences and consumed more than 300,000 tokens in a day without any further human prompting, The Register reports. The agent had read-only permissions and no connected email or chat apps, yet spun up subagents to survey systems. Geordie’s chief AI officer said personal agents set many of their own tasks.
Cerebras’ Feldman and Michael Jordan push back on AI doom talk
At TechCrunch Disrupt, Cerebras CEO Andrew Feldman said “some of the leaders in the AI world maybe shouldn’t be allowed to talk in public,” arguing risk talk drowns out benefits, though he said AI should be regulated reasonably; he also predicted home AI assistants that run partly on-device. UC Berkeley’s Michael I. Jordan called superintelligence warnings science fiction and AI firms’ calls for regulation “theater,” The Decoder reports. On the All-In Podcast, David Sacks criticized what he called “rotating hoaxes” around AI.
Nvidia plans investments in Doctolib, Pigment and two other French startups
Nvidia announced planned investments in four French startups: Doctolib, Entalpic, Genesis AI and Pigment. The company says it has committed nearly $10 billion to Europe’s technology ecosystem, and Sifted reports CEO Jensen Huang said the figure covers the past year. Nvidia did not disclose the size of each new investment.
All-In hosts: AI is ending coding as we know it and devaluing software IP
In clips posted by the All-In Podcast, Jason Calacanis argued that AI is the end of coding as he knows it, with developers handing work to agents rather than writing code together. In another clip, Chamath Palihapitiya argued that software intellectual property has become worthless, pointing to open-source tools presented as alternatives to Adobe’s main products. Both are opinions from the hosts.
NHTSA reviews petition alleging Tesla robotaxis speed in school zones
The National Highway Traffic Safety Administration is evaluating a petition, received September 23, alleging that Tesla’s model-year 2026 robotaxis, the Model Y and Cybercab, exceed posted speed limits in active school zones, The Verge reports. Tesla runs about 1,000 robotaxis. Waymo has faced similar inquiries, including over passing school buses.
Science Corp. implants PRIMA retinal chip in first Stargardt patient
Science Corporation CEO Max Hodak, a Neuralink co-founder, said at TechCrunch Disrupt that the first trial patient with Stargardt disease received its PRIMA retinal implant in May in Sydney. The company says the patient has been able to read a letter and then a word during rehabilitation. It is a single-patient trial result, not an approval for the condition.
Facebook gets Instagram’s LLM-driven “Your Algorithm” controls
Meta is bringing Instagram’s “Your Algorithm” feature to Facebook and Facebook Reels, letting users see and adjust the topics shaping their feeds. Meta says it is using large language models more in its ranking systems.
MIT Technology Review: AI labs are buying data from dead startups
AI labs, training-data vendors and large tech companies are buying code bases and workspace data such as emails, chats and documents from startups that are shutting down, MIT Technology Review reports. SimpleClosure says its Asset Hub has handled more than 150 deals and expects to return over $10 million to startups by the end of 2026; 47% of its deals now include code or data sales.
Shing-Tung Yau launches AI-math alliance as Western mathematicians voice doubts
Fields Medalist Shing-Tung Yau launched an alliance of eight institutions in Beijing that aims to build a network of more than 1,000 researchers to push AI into mathematics, The Decoder reports, citing the South China Morning Post. Yau warned that China is falling behind the U.S. Meanwhile, 25 Fields Medalists, including Terence Tao and Peter Scholze, have raised concerns that AI-generated proofs erode mathematical understanding.
Ghostscript flaw could let attackers take over servers through PDF uploads
A critical heap buffer overflow in Ghostscript, CVE-2026-39919, could allow remote code execution through crafted PDF files on systems that process uploads with the library, BleepingComputer reports. Artifex fixed it in version 10.08.0, released September 8, and the report says no exploitation has been observed yet. Exposure depends on how a service handles uploads.
Liquid AI releases LFM2.5-24B-A2B, a 2B-active model for on-device agents
Liquid AI released LFM2.5-24B-A2B, a reasoning model with 24 billion total parameters, 2 billion active per token and a 131K-token context window, aimed at private agents on local hardware. The company reports 184 tokens per second on an Apple M5 Max and up to 21,000 output tokens per second on a single Nvidia B300; the benchmarks are its own.
GitHub swaps an LLM pipeline for a small model in secret scanning
GitHub says its secret scanning now pairs regex candidate matching with a fine-tuned ModernBERT classifier instead of remote LLM calls. The company reports an 11.5% rise in owner-weighted alert acceptance, an 86% cut in median post-push scan time and more than 98% lower serving cost.
Nvidia readies apps for RTX Spark laptops arriving October 16
Nvidia says RTX Spark laptops, built on a superchip with up to 128GB of unified memory and 1 petaflop of AI compute, go on sale Friday, October 16. Ahead of launch it listed updates to agent apps including GitHub Copilot and OpenClaw, and said PyTorch, CUDA and its libraries are ready for the platform.