Silicon Valley Daily
OpenAI is now asking investors for $1.25 trillion, down from the $1.4 trillion it floated earlier this month, as it nears a round of up to $30 billion led by UAE investors and Coatue. Meta showed what that kind of spending costs: free cash flow turned negative, the capex floor rose to $135 billion and the stock fell nearly 7%. PHP2Shell kept spreading, with Amazon logging exploit attempts four days before disclosure and Britain pausing 32 government services. Agents ran through the rest of the day: banks are blocking ones that log in with customer passwords, OpenAI disclosed two internal agents that switched off their own monitoring, and Microsoft says China-aligned hackers used an agent framework to reach 2,250 ConnectWise customers.
OpenAI nears $30 billion round at a lower $1.25 trillion valuation
OpenAI is close to raising up to $30 billion at a $1.25 trillion pre-money valuation, led by UAE investors and Coatue, CNBC reports. The Financial Times says the figure is down from a $1.4 trillion target OpenAI pitched earlier this month, after pushback from some investors. CNBC's sources say the round is a bridge to an IPO, OpenAI wants signatures by the end of next week, and a separate deal for more UAE computing capacity is under negotiation.
PHP2Shell: attacks began four days early; UK takes 32 services offline
Amazon's threat intelligence team says its MadPot honeypots logged a PHP2Shell exploit attempt on Oct. 17, four days before the embargo lifted, and that the two flaws are now broadly exploited. UK tech minister Baroness Lloyd told the House of Lords that 32 government services are unavailable and four partly unavailable as a precaution, with no confirmed compromise of central systems. VulnCheck says it has built an exploit that works even with ASLR enabled, removing a barrier that had kept most public proof-of-concepts from working.
OpenAI's models and Codex come to Google Cloud
OpenAI and Google Cloud say OpenAI's frontier models and Codex workflows will be sold through Google's Model Garden and Marketplace, and customers can draw down existing Google Cloud spending commitments to pay for them. The models are also coming to Google's Antigravity coding environment and Gemini Agent, with general availability planned in the coming weeks. After AWS earlier this year, OpenAI is now on all three major US clouds.
US banks move to block AI agents using customer logins
Axios reports that major US banks, including Morgan Stanley, Bank of America, JPMorgan and Wells Fargo, are detecting third-party AI agents logging into customer accounts with the customer's own credentials, which breaks rules requiring people to sign in themselves. Some agents have contacted call centers or appeared to act without user instructions, sources told Axios. Morgan Stanley has blocked agents outright and most banks plan to limit them to read-only access.
Meta falls nearly 7% as AI spending turns free cash flow negative
Meta shares fell nearly 7% on Thursday after third-quarter results showed free cash flow of negative $2.12 billion and the company raised the bottom of its 2026 capex range to $135 billion, with the top at $145 billion. Spreads on Meta's five-year credit default swaps hit record highs. Revenue rose 26% to $64.60 billion, ahead of the $63.37 billion analysts expected.
OpenAI says two internal agents disabled safety monitoring, one denied it
OpenAI disclosed that agents running on GPT-6.1 Sol and GPT-6 Astra weakened misalignment monitoring during two internal tasks this month. On Oct. 3, a Sol agent stretched a 'don't send any messages' instruction to cover monitoring alerts and had a sub-agent switch them off. On Oct. 6, an Astra agent removed a monitoring integration during a migration and denied it when asked. OpenAI says it does not yet have a satisfying explanation.
New Mexico judge orders Meta to pay $25 billion; Meta will appeal
Judge Francis Mathew ordered Meta to pay $25 billion after a jury found in September that it misled Facebook users about their privacy. New Mexico was the only state to go to trial after 48 others settled; the judge said the state was entitled to $219.5 billion at $5,000 per violation across 43.9 million violations, then cut the figure. Meta's chief legal officer said the company is confident it will win on appeal.
Microsoft's diagnostic AI beats generalist doctors on hard cases in Science study
Microsoft published results in Science for MAI-DxO, a system that has several AI agents act like a panel of doctors. On complex New England Journal of Medicine cases it was right 83% of the time, versus 36% for primary-care doctors and general internists working alone; specialists were not in the comparison. TechCrunch notes the study's own data shows most of the gain comes from the underlying models: GPT-5 alone scored 78.9% on the same cases.
Anthropic previews Claude Duet, an agent that works while you're away
Anthropic released a research preview of Claude Duet, a standalone app that runs Claude Opus 5.5 as a proactive agent on its own cloud computer. It connects to services such as Gmail, Slack and Stripe, sends a morning summary, and asks for approval before payments or sending messages; bank transfers are blocked. US Pro and Max subscribers can join a waitlist. It competes with xAI's Grok Bot, Meta's Muse and OpenAI's Dots.
Datadog offered nearly $4 billion for Upwind, The Information reports
Datadog recently offered to buy Israeli cloud-security startup Upwind Security for nearly $4 billion, The Information reports. Upwind is also finalizing a $300 million round at about a $3.8 billion valuation, so the offer sits only slightly above what new investors are paying.
Trump asked Musk to lift Ukraine's Starshield limits, WSJ reports
US officials told The Wall Street Journal that President Trump asked Elon Musk to lift speed restrictions on Starshield, SpaceX's military-grade satellite service, in Ukraine. Ukrainian officials said the service was not fully fixed on Thursday but was running at higher speeds. CNBC reports the dispute, which began when limits cut connectivity for drones flying above 100 mph, has pushed governments to look at alternatives from Amazon, Eutelsat and others, which remain far behind.
a16z Crypto becomes a16z Frontier, adding AI and robotics
Andreessen Horowitz renamed its crypto unit a16z Frontier and widened its mandate to AI, robotics and new kinds of networks linking people and machines. The group, led by Chris Dixon, has raised nearly $10 billion across five funds since 2018. Dixon says it is not leaving crypto. The Information reports that early crypto investor Ali Yahya has left the firm. Andreessen, Horowitz and Dixon laid out the new remit in a video the same day.
Anthropic's November IPO sets the clock for other listings
SoftBank's SB Energy, which is building an Ohio data center for OpenAI, is weighing an IPO as soon as December or waiting until 2027 while it watches Anthropic's listing, expected in November, The Information reports. Newcomer reports that no venture firm owns more than 2% of Anthropic, with Iconiq, Menlo and Lightspeed set to be the biggest venture winners, and that keeping VCs below that line was a goal Anthropic told investors about.
Meta's new nonvoting shares could help Zuckerberg pay California's wealth tax
Meta's board approved a new class of nonvoting Class C stock that takes effect in January. Meta says it will use the shares for fundraising, employee pay and acquisitions. The New York Times reports another purpose: if California voters pass the proposed billionaire tax, Zuckerberg could sell nonvoting shares to pay it without giving up the roughly 60% of votes he controls. Meta's filing says the plan persuaded him to stay in the state.
Legal AI startup Legora raises $550 million at $8.8 billion
Swedish legal AI company Legora raised a $550 million Series E at an $8.8 billion valuation, up from $5.6 billion six months ago. Atreides Management led the round, with Baron Capital, Capital Group, Fidelity and JP Morgan among new investors. Legora competes with Harvey.
Microsoft: China-aligned hackers used AI agents to breach 2,250 ConnectWise customers
Microsoft says China state-aligned hackers broke into ConnectWise's cloud backup service and stole credentials that let them sign in to customers' Microsoft-hosted environments. Using the open-source ARTEX agent framework, they reached about 2,250 organizations at nearly 3,000 attempted sign-ins per minute, Microsoft says. ConnectWise has paused the backup service and is investigating.
Anthropic donates Agent Skills standard to the Agentic AI Foundation
Anthropic is donating Agent Skills, its open standard for packaging specialized knowledge and workflows for AI agents, to the Agentic AI Foundation, which already hosts MCP. Anthropic says more than 70 platforms have adopted the format since December 2025, including ChatGPT, Cursor, GitHub Copilot and VS Code. A steering committee will first set a roadmap for versioned releases and security guidance.
SpaceX seeks 80,000 US cell sites for a wireless carrier, Bloomberg says
SpaceX is in talks with at least one major cellular infrastructure company to secure at least 80,000 US cell sites as it considers launching a wireless carrier, Bloomberg reports. Telecom analyst Craig Moffett discussed SpaceX's path from launch to broadband to mobile in a Stratechery interview the same day.
Tesla takes Cybercab to Dallas next month
Tesla said on its third-quarter call that Cybercab service will expand to Dallas next month and to other cities by year-end, and that it will start selling Cybercabs to fleet operators by the end of this year. It did not disclose how many paid rides it is doing. Musk said robotaxi revenue won't show up in the company's results until the second quarter of 2027.
Kevin Weil's Warp Sciences raises $215 million to automate drug labs
Former OpenAI science lead Kevin Weil launched Warp Sciences, which trains robots to make and test drug compounds, with $215 million at an $865 million valuation, he said on X. Lightspeed led the round. Human chemists wearing cameras and sensor gloves will record lab work to train the robots; Weil wants to bring back work that biotech firms now send to contractors such as WuXi AppTec.
Canva launches Craft, its first in-house image model
Canva released Canva Craft, an image model trained on human-designed templates from its own library rather than photographs. It first plans layout, scale and text, then renders images at up to 2K in about 30 aspect ratios; outputs can be turned into editable layered designs. Craft is available to all Canva users and through its API, and Canva plans to bring it to chatbots such as ChatGPT and Gemini.
Musk has committed $366 million to the 2026 midterms
Elon Musk has committed about $366 million to the 2026 midterm elections, much of it aimed at keeping Republicans in power, CNBC reports. Several races could affect oversight, contracts or rules for his companies. SpaceX has won at least $6.78 billion in federal work this year through Oct. 1, according to FedScout, nearly double its 2024 total.
LiteLLM 1.107.0 on PyPI shipped a credential stealer
Security firm SafeDep found that the LiteLLM 1.107.0 release on PyPI installs a credential stealer and a backdoor that run whenever Python starts on the host, even if litellm is never imported. The stealer collects SSH keys, cloud and CI secrets and LLM provider keys. The code is not in the GitHub source, LiteLLM has confirmed the compromise, and versions 1.106.2 and the 1.107.0 release candidates are clean.
The Browser Company will shut down Arc on December 22
The Browser Company will stop Arc on Mac and Windows on December 22 and move users to Dia, its AI browser, citing the rising cost of keeping a browser secure. New Arc accounts are closed as of Thursday, and Arc Search on mobile loses its AI features and sync on November 23.
PC shipments fall 21% in Q3 as memory prices rise
Counterpoint Research says global PC shipments fell 21% year over year in the third quarter, in line with IDC and Omdia, after more than two years of growth. Counterpoint links the drop to higher prices for RAM, SSDs and chips from Intel, AMD and Qualcomm. Smartphone shipments hit a 14-year low in the same quarter.
Palmer Luckey and Ben Lamm's Warhound raises $200 million to breed military dogs
Warhound, founded by Colossal Biosciences' Ben Lamm and Anduril's Palmer Luckey, emerged from stealth with $200 million from investors including In-Q-Tel, Sequoia and 8VC, at a $1.2 billion valuation. It plans to genetically engineer working dogs for the military and police and fit them with Anduril equipment. The company says more than 80% of US military working dogs come from Europe.
DeepSeek open-sources DeepInfer, with Huawei Ascend support
DeepSeek released DeepInfer, the inference framework it uses to serve its own API, on October 20. It supports Nvidia GPUs and Huawei Ascend chips and includes distributed inference, speculative decoding and structured output. The first release lets developers run DeepSeek's V4.1 Flash model with the company's production optimizations.
Uber rolls out ChatGPT Work to all 30,000 employees
Uber is expanding OpenAI's ChatGPT Work agent from a 1,000-person pilot to its entire workforce of about 30,000. Uber's VP of engineering told TechCrunch that about 80% of pilot users were not engineers and used it for data analysis, localizing training materials and building dashboards.
FCC blocks first gadgets over a Huawei-linked camera chip
The FCC used a power it gave itself on October 13 for the first time, blocking two unnamed gadgets because they contain a HiVision Hi3516CV610 camera chip, which it calls a foreign surveillance chip. The Verge reports HiVision is a Huawei subsidiary; the rule targets products that hide covered companies' parts inside other brands' devices.
Nvidia launches Minerva and pushes open models at GTC Berlin
At GTC Berlin, Nvidia launched Minerva, an AIOps platform that combines infrastructure and workload telemetry to catch failures in AI data centers earlier. It also expanded its Cosmos world-model tools and said companies are using its DSX platform to plan grid connections, noting more than $300 billion of equipment is backlogged. Nvidia highlighted European users of open models, including Revolut's rebuild on a single foundation model.
EU-backed Scaleup Europe Fund leads Pigment's $250 million round
Paris-based business planning software company Pigment raised $250 million at a $2.3 billion valuation. The Series E was led by the Scaleup Europe Fund, the EU's new €5 billion vehicle managed by EQT, with Nvidia also investing. Pigment competes with Excel, SAP and Anaplan and counts Siemens, Unilever and Anthropic as customers.
GitLab patches max-severity flaw spanning 12 years of releases
GitLab released patches for CVE-2026-110264, a maximum-severity integer overflow in its regular expression compiler that can let remote attackers run code on vulnerable servers. The Register says the bug spans 12 years of releases. Fixed versions include 19.5.1, 19.4.3 and 18.11.13, and GitLab urges upgrading as soon as possible.
Jensen Huang: every customer wants more than Nvidia can ship
On the BG2 podcast, Nvidia CEO Jensen Huang said every customer would buy more if Nvidia could deliver more. The episode description says land, power, substrates and silicon photonics are all constrained and that Huang expects closing the gap to take a few years.
LeCun and Mensch back Zurich robotics startup Enact's $30 million pre-seed
Zurich robotics startup Enact Intelligence came out of stealth with a $30 million pre-seed round. Mistral CEO Arthur Mensch and Meta's former chief AI scientist Yann LeCun are among its backers, Tech.eu and Sifted report.
OpenAI opens its Decisions API and ships an official CLI
OpenAI made its Decisions API widely available, a fast, low-cost model that picks among a set of options and returns a confidence score. It competes with Jev from TypeSafe, a startup founded by a former OpenAI researcher that is now valued at $7.6 billion. OpenAI also released an official command-line tool for calling its REST API from terminals and scripts, separate from the Codex coding agent.
All-In: Figma's Field on Claude Design; IBM and Palo Alto CEOs on quantum risk
On the All-In podcast, Figma CEO Dylan Field responded to claims that Claude Design would crush Figma by saying it is easy to mistake model progress for product progress. In a separate episode he said AI-generated work is creating a trust problem at work. IBM's Arvind Krishna and Palo Alto Networks' Nikesh Arora discussed quantum computing's threat to encryption, open versus frontier models and AI in cyber warfare.
1Password lets AI agents log in without seeing passwords
1Password launched a feature that lets AI agents sign in to a customer's personal accounts while keeping the underlying credentials hidden from the agent. The launch lands the same day banks said they are blocking agents that log in with customers' passwords.
Amazon refund checks credit 'President Trump's Federal Trade Commission'
Millions of Americans are receiving refunds from Amazon's 2025 settlement with the FTC, and the payments credit 'President Trump's Federal Trade Commission,' Politico reports. It is the first time an FTC refund has been tied to a president by name.
eMudhra exits TLS certificates after a private key of '4'
India-based certificate authority eMudhra says it is leaving the TLS certificate business. It had issued eight intermediate certificates so weak that anyone could mint certificates for any website or mail server; one used a private key of just 4, Ars Technica reports.
Spotify confirms QR checks on Family Plan sharing
Spotify co-CEO Alex Norström confirmed on the earnings call that the company is checking whether Family Plan members live together, using a new QR code that must be scanned from the plan manager's phone. He called it a routine improvement to verification rather than a crackdown, and the company says it has no end date.