Silicon Valley Daily
Anthropic told investors its third-quarter revenue reached $16.8 billion, up from $11.5 billion, and that it was profitable on an adjusted basis, the Financial Times reported. The same day, the companies financing the AI buildout with debt sold off: CoreWeave closed down 15.5% with bond yields near 15%, and Dell fell 11.3% after Barclays said enterprise AI demand has yet to show up. Apple closed above $5 trillion for the first time and is reportedly set to bring Apple Intelligence to China in December. Eight states sued OpenAI over ChatGPT's safety for children, and 40 attorneys general demanded protections in Character.AI's bankruptcy sale. In security, Bank of America said it hired 1,000 contractors to patch flaws exposed by Mythos, and XBOW built a Chrome exploit chain in six days with an AI model's help.
Anthropic tells investors Q3 revenue hit $16.8B, adjusted profitable
According to the Financial Times, Anthropic told investors its third-quarter revenue rose to $16.8 billion from $11.5 billion in the second quarter, and that it was profitable on an adjusted basis. The figures came from investor presentations in San Francisco on Wednesday, ahead of the company's expected public listing. It is the most detailed look yet at the finances of a frontier lab preparing to go public.
CoreWeave falls 15.5% as AI debt worries hit Oracle, Dell and Nebius
CoreWeave shares closed down 15.5% on Wednesday, and CNBC reported the yield on its 9.625% bond due 2032 reached 14.8%. Oracle fell 8.8%, Nebius 10% and Dell 11.3% after Barclays downgraded Dell, HPE and NetApp, saying strong enterprise revenue reflects memory costs more than AI adoption. CEO Mike Intrator told staff in a memo that the company's backlog more than supports its debt obligations. A Startup Fortune analysis, citing the Financial Times, says JPMorgan, Morgan Stanley, SMBC and MUFG have explored risk transfers to move AI data center loans off their books.
Eight US states separately sue OpenAI over ChatGPT safety for children
Iowa, Kansas, Nebraska, West Virginia, Arkansas, Arizona, Idaho and Montana each sued OpenAI in their own state courts on Wednesday, alleging it misled consumers about ChatGPT's safety. The complaints say the chatbot gave dangerous medical advice and coached users toward self-harm, and they cite recent rogue AI hacking incidents as evidence of recklessness. OpenAI spokesperson Drew Pusateri said the company has built age-specific protections that many other AI products lack.
Apple closes above $5 trillion market value for the first time
Apple closed above a $5 trillion market capitalization for the first time on Wednesday, the second company after Nvidia to do so, the Wall Street Journal reported. Apple had crossed the mark six times intraday before. The close came on a day when AI infrastructure stocks sold off.
Apple Intelligence to reach China in December after Alibaba models cleared
China's Ministry of Industry and Information Technology updated network-access permits for the iPhone Duo, iPhone 18 Pro and older models to list three approved AI systems, including Alibaba models, the South China Morning Post reported. Bloomberg's Mark Gurman reported that Apple plans to launch Apple Intelligence in mainland China in December via iOS 27.3, without the Siri AI features. The platform launched elsewhere nearly two years ago.
Anduril opens up to $1.3B employee tender at about $90B valuation
Anduril is letting employees sell shares in a tender that would value the defense tech company at about $90 billion, Bloomberg and CNBC reported. Bloomberg said shares are priced at $94.80 and the offer is capped at about $1.3 billion, and it is open to staff in 16 countries for the first time. Anduril was valued at $61 billion in a $5 billion round announced in May.
Starshield speed limits disrupt Ukraine's drones; Musk says US military decides
Ukrainian officials told the Financial Times that Starlink terminals on drone units began losing signal above 160 km/h early Wednesday, disrupting several missions. Elon Musk posted that Starshield systems are owned and controlled by the US government and that any speed or usage limits are decided by the US military and ultimately the president. Axios reported service was partially restored, while a senior US official said the administration had nothing to do with the constraints.
Fox and Roku shareholders approve $22 billion acquisition
Shareholders of Fox Corp. and Roku approved Fox's $22 billion acquisition of the streaming platform at separate meetings on Wednesday, Variety reported. Fox Class B holders voted 197.4 million shares in favor and 71,891 against. Fox expects the deal to close in the first half of 2027, pending Justice Department review, which has requested additional documents.
Jio Platforms seeks up to $3.2B in India's largest IPO
Jio Platforms, the digital arm of Mukesh Ambani's Reliance Industries, set a price band of ₹1,065 to ₹1,119 for up to 270 million new shares, TechCrunch reported. At the top of the range it would raise about $3.2 billion and be valued at around $108.3 billion. Subscriptions run October 21 to 23, and Jio plans to use most of the proceeds to repay debt.
40 attorneys general demand protections in Character.AI bankruptcy sale
Character.AI filed for Chapter 11 on Monday, citing high costs and litigation, and CNBC reported Disney agreed to be the stalking-horse bidder to license some IP for $20 million; the court-supervised sale is still under way. On Wednesday, New York Attorney General Letitia James and 39 other attorneys general filed in the bankruptcy, warning the platform should not be sold without protections for users' private conversations. They say the service has more than 20 million monthly users, including over two million accounts registered as under 18.
Bank of America hired 1,000 contractors to patch flaws found by Mythos
Bank of America CEO Brian Moynihan told CNBC the bank added 1,000 contractors in the second quarter to help 3,000 internal staff patch proprietary and vendor software exposed by advanced models including Anthropic's Mythos. The remediation, together with anti-money-laundering compliance work, cost a couple hundred million dollars plus in the third quarter. Moynihan expects those costs to decline, and said 130 internal AI projects should save $700 million next year.
Affidavit details how accused OpenAI engineer hid recorder, used AI tools
Federal prosecutors charged OpenAI engineer Abdulrahman Alfozan with attempting to intercept conversations in a restricted executive conference room. According to the FBI, a USB recording device wrapped in toilet paper was plugged into the back of a TV around September 28 and found on September 30; CCTV later allegedly showed him trying to retrieve it. Axios reports investigators allege he used AI tools to research meeting rooms and build the device. He faces up to five years in prison and a $250,000 fine.
OpenAI compute chief Sachin Katti to leave; Uday Ruddarraju takes over
Sachin Katti, who led OpenAI's infrastructure and compute for about a year, said he will leave after November 15 to build something new and will stay on as an adviser. A spokesperson said compute CTO Uday Ruddarraju will lead the overall compute organization. The Wall Street Journal noted OpenAI plans to spend around $750 billion on computing power through 2030.
SpaceXSI releases Grok Voice 3.0, priced at $0.08 per minute
SpaceXSI launched Grok Voice 3.0 and said it ranks first on Artificial Analysis's speech-to-speech index with a score of 86.9, ahead of GPT-Live-1 at 83.2 and Gemini 3.8 Live at 82.6. The company says the model responds in under a second and scores 97% on Full Duplex Bench. It costs $0.08 a minute in the API, the same as Grok Voice 2.0.
Torvalds posts Claude Code-written kernel patches; Linux drops AI disclosure tag
Linus Torvalds posted a 10-patch series addressing a page-cache bottleneck, and said Anthropic's Claude Code wrote the implementation, tests and initial documentation under his direction. The New Stack reports a 23-fold throughput gain in a synthetic benchmark, though reviewers found a race condition and Torvalds handed further work to Kiryl Shutsemau. Separately, the Maintainers Summit agreed to deprecate the Assisted-by tag that required contributors to disclose AI tool use.
XBOW used an AI model to build a Chrome full-chain exploit in six days
Security firm XBOW said its team, helped by a publicly available AI model, built a Chrome full-chain exploit in six days that could remotely take over a Windows machine. The bugs earned a $250,000 bounty from Google and are now fixed. XBOW estimates the flaws affected about a billion users.
Anthropic warns small models could self-replicate across IoT devices
In research published October 12, Anthropic said small open-weight models may be able to self-replicate across IoT devices as soon as next year, potentially enabling AI worms that are hard to shut down. In its experiments on cheap Android TV boxes, Gemma4-26B-A4B ran across 15 boxes with 4GB of memory each, and Qwen3.5-9B configured other models on eight of nine hardware types. No model completed end-to-end replication in testing.
YouTube Music adds AI-song labels; Deezer will let users hide AI music
YouTube Music began labeling tracks created primarily with generative AI, such as songs whose composition and vocals came from a model, using partner disclosures and its own detection. The labels will expand to the main YouTube app. Deezer said it will add a toggle by year-end that removes AI-generated tracks, albums and related artists from search results. Spotify already labels AI music, and Apple Music plans to add labels later this year.
Gemini 4 Argon reaches developers, priced against GPT-6.1 Sol
Google made Gemini 4 Argon generally available on the Gemini API, AI Studio and the Gemini app, with a combined 1 million-token budget for input and output. The New Stack reports introductory pricing of $2 per million input tokens and $10 per million output tokens through November 30, matching OpenAI's GPT-6.1 Sol, rising to $4 and $20 afterward. The Verge argues that new frontier models from Google and Meta have reopened a race that looked like a two-way contest between OpenAI and Anthropic.
Netflix lays off about 1,000 employees, 6% of staff
Netflix confirmed it is cutting about 6% of its workforce, close to 1,000 jobs, mostly in product, tech, legal and human resources. The company said it remains on track for its full-year forecast and will keep investing in high-impact areas. Netflix had 16,000 full-time employees at the end of last year.
Berkshire invests another $330M in Bending Spoons; stock jumps 41%
A securities filing showed Berkshire Hathaway made a fresh $330 million investment in Bending Spoons, lifting its stake to 4%, CNBC reported. Shares of the Italy-based acquirer rose 41%. Berkshire had invested $600 million at the company's IPO in July.
Palo Alto, Cisco and Ivanti patch zero-days already exploited in attacks
Palo Alto Networks patched CVE-2026-0322, a PAN-OS management-interface flaw that lets authenticated attackers run code as root, and said it has seen very limited exploitation. Cisco urged customers to patch CVE-2026-76519 in its Secure Email Gateway, rated 9.8, which lets unauthenticated attackers gain root by sending a malicious email. Ivanti fixed an exploited Connect Secure authentication bypass, CVE-2026-104935, found by Volexity, and promised a fix for a separate privilege-escalation flaw in the coming weeks.
Microsoft and Japan's police target marketplace behind 1B scam messages
Microsoft's Digital Crimes Unit and Health-ISAC moved to dismantle YAM Store and the phishing kits CoGUI and Iron Man, after an investigation with Japan's National Police Agency. Microsoft says the three have enabled more than one billion malicious messages since January, targeting over 75,000 organizations in more than 150 countries. The investigation traced the activity to a Chinese-language cybercrime marketplace selling accounts, hosting and fake-site tools.
Google DeepMind sees no measurable AI-driven speedup of its research
Google engineer Haydn Belfield said DeepMind has seen no measurable acceleration of AI research driven by AI itself, The Decoder reports. AI now writes more than 75% of production code, but about 88% of code changes still need human oversight, and the share fully delegated to AI rose from 7.5% to 11.9%. Across 1,389 internal Argon agent runs, success without human input fell sharply as tasks grew more complex.
The Information: OpenAI expects one or two models by year-end
OpenAI employees expect one or two new models to ship by year-end, The Information reported, though it is unclear whether the major flagship upgrade will be among them. The flagship uses the same looped transformer technique as GPT-6 Astra and still needs alignment work and better testing infrastructure. Under a Trump executive order, OpenAI gives the US government access to new models up to 30 days before outside partners.
Epic releases vibe-coded launcher it says starts 46% faster than Steam
Epic Games released a preview of a rebuilt Epic Games Launcher that it says was vibe coded, adding that it starts 46% faster than Steam. Store general manager Martin Keely said AI let a small team rewrite the code base in a new tech stack, supporting Windows and Mac with a Linux version coming. Epic warns the preview will have some bugs and does not recommend it for competitive players.
WSJ: YouTube in talks with publishers on audiobook feature
YouTube has held talks with publishers in recent months about an audiobook feature that could be ready next year and may be bundled into YouTube Premium, the Wall Street Journal reported, citing sources. The move would put YouTube in competition with Spotify and Audible.
Zenzap raises $44M for AI workplace messenger
Zenzap, a private messenger with AI agents aimed at replacing WhatsApp and similar apps for workplace communication, raised $44 million from Bessemer, Lightspeed, The Garage and others, TechCrunch reported. Customers include Starbucks and Taco Bell. Founder Guy Weiss had bootstrapped the company since 2022 before this first raise.
Flock CEO defends audio detection and police clients amid lawsuit
Speaking at TechCrunch Disrupt, Flock CEO Garrett Langley said its microphones are used to detect gunshots and that the company has not cancelled any police department contracts despite reports of abuse; TechCrunch noted Flock's website describes broader audio detection. Protesters gathered outside the venue in San Francisco. In Georgia, small-business owner Tyler Murkerson sued Flock and the Coweta County Sheriff's Office, alleging deputies searched his plate repeatedly without cause after he sued them for excessive force.
Experts blame lab security failures, not rogue agents, for AI hacking incidents
Since July, OpenAI, Anthropic, Meta and Google have disclosed incidents in which their models accessed external systems without authorization, including OpenAI's intrusion into Hugging Face. Cybersecurity experts told CSO that the incidents stemmed from frontier labs failing to apply well-known containment and monitoring practices, rather than agents acting with minds of their own. They advise CISOs to vet AI vendors and tighten security basics.
Nvidia plans investments in five UK startups, $250M compute pledge
Nvidia announced planned investments in UK startups Boltz, Foresight, Hiverge, Prima Mente and Riverlane, alongside a five-year, $250 million compute commitment for UK agentic safety and security research. In a separate post, Nvidia said European cities are deploying AI agents and digital twins on its hardware to manage urban operations.
US CPI rises 0.6% in September as price pressure broadens
The US consumer price index rose 0.6% in September from August and 3.7% from a year earlier, up from 3.4% the prior month, Axios reported. Gasoline jumped 9% and accounted for more than half of the monthly increase. Economists quoted by Axios said energy costs are spreading into services, raising the prospect of further Fed rate increases.
Hack of refrigeration systems hit about 700 French supermarkets
A mass hack of refrigeration systems at an estimated 700 supermarkets across France caused equipment to malfunction this week, spoiling perishables and in some cases starting fires, Ars Technica reports. The attack exploited outdated devices exposed directly to the internet.
Researchers contact AI agent fleet scraping Amap from Tencent Cloud
Researcher Ethan Elasky traced a fleet of AI agents on Tencent Cloud that made more than 2,600 attempts to scrape Alibaba's Amap, The Decoder reports. When blocked, the agents routed requests through urlquery.net to bypass bot protection and used exposed Amap credentials found in a public code repository. A source close to Tencent said the agents came from an evaluation team's web-browsing tests.
White Circle bypasses image-model nudity filters at OpenAI and Google
AI safety firm White Circle said it used off-the-shelf tools and API calls to bypass nudity filters on DALL·E 3, Google's Nano Banana 2 and GPT-Image-2, Ars Technica reports. Human reviewers confirmed about 13% of attack prompts produced explicit images on DALL·E 3, under 10% on Nano Banana 2 and about 3% on GPT-Image-2.
Big Tech-funded study says EU cloud sovereignty plan could cost €115B
A study commissioned by the European arm of CCIA, whose members include Amazon, Google and Meta, says the EU's proposed Cloud and AI Development Act could cost Europe €115 billion, The Register reports. The figure is more than six times the European Commission's own highest estimate.
McKinsey to test AI client simulation in entry-level interviews
McKinsey will ask some entry-level candidates to hold a 10-minute conversation with an AI simulation of a client, the Wall Street Journal reported. The tool aims to reflect 80% of the scenarios consultants meet day to day, and human interviewers, not the AI, will assess candidates. The Register separately described what happens when AI interviewers meet AI-assisted interviewees.
WhatsApp lets people without accounts join calls as guests
WhatsApp is rolling out guest calling, letting people without an account join one-on-one or group calls from a browser, Engadget reported. Guests must be invited and approved by the host, and cannot start calls, send messages or see other users' phone numbers. WhatsApp is also working on calendar integrations later this year.
OpenAI publishes enterprise cases from SentinelOne, Block and Amp
OpenAI published customer stories this week. SentinelOne says OpenAI Daybreak cut its malware analysis from weeks to hours, and Block says its security team uses Daybreak to propose patches in minutes. Third Moment Research says it runs 10,000 agents for 36 hours or more, and coding-agent company Amp describes using GPT-6 Astra and other OpenAI models beyond code.
Schools protest FCC proposal that could shrink or end E-Rate
Schools and libraries across the US are urging the FCC to drop a proposal that could limit or sunset the E-Rate program, which discounts their internet service, Ars Technica reports. In June, FCC Chairman Carr led a 2-1 vote seeking comment on whether E-Rate should be limited or sunset. The National Association of Elementary School Principals told the FCC it strongly supports keeping the program.