Silicon Valley Daily
Binance reopened withdrawals on Sunday and saw $3.64 billion in net outflows in the first 9.5 hours, 2.1% of its assets, after a $6.3 billion theft the FBI attributes to North Korea. Two days before the midterms, data centers are on the ballot: candidates backing a moratorium could win up to 15 of 36 governor's races, and OpenAI is preparing a post-election push for AI 'speed limits' while renaming its facilities 'compute campuses.' Saudi Aramco is putting about $2.6 billion in cash into Humain for 49%, and Elon Musk is adding two more sites to Colossus 2. Markets are pricing the buildout's risk: default insurance on the five biggest AI spenders costs the most since 2021. Inside the companies, Anthropic says its internal Claude use at list price now exceeds its cash payroll, and Linus Torvalds says AI-found fixes pushed Linux 7.4 to nearly 20,000 commits.
Binance reopens withdrawals; $3.64B leaves in the first 9.5 hours
Binance began restoring withdrawals in phases on Sunday after saying it had covered most of the $6.3 billion stolen in the largest crypto heist on record. The exchange told The Block that net outflows reached $3.64 billion, or 2.1% of its assets, in the first 9.5 hours, and that volumes eased after an initial spike; DefiLlama put 24-hour net outflows at $4.05 billion. The FBI has attributed the theft to North Korea's TraderTraitor group, and Bloomberg reports Binance ran a global war room of about 100 people coordinating with other crypto firms and the FBI.
OpenAI readies post-midterm push: AI 'speed limits' and 'compute campuses'
Axios reports OpenAI will launch a policy and PR campaign right after Tuesday's midterms, timed to the lame-duck session that opens Nov. 9. Chief global affairs officer Chris Lehane's team has stockpiled proposals since August, including mandatory safeguards for the most powerful AI systems that OpenAI calls speed limits. An OpenAI spokesperson confirmed to Business Insider that the company will call data centers 'compute campuses,' saying the term better describes the scale and role of the infrastructure; a Pew poll cited by BI found 54% of Americans say AI data centers are bad for the environment.
Saudi Aramco buys 49% of AI company Humain for about $2.6 billion
Saudi Aramco has taken a 49% stake in Humain, the AI company launched in May 2025, through a wholly owned subsidiary, GamesBeat reports; PIF keeps 51%. Aramco will contribute about SAR 9.9 billion ($2.6 billion) in cash in 2026, plus AI, compute and digital assets. Humain is building a full-stack AI business spanning data centers, compute, models and applications, and Aramco CEO Amin Nasser said the company has realized $11.3 billion in technology value since 2023, much of it from AI.
Data center moratorium backers could win up to 15 governor races Tuesday
An analysis by The Information finds candidates backing a data center moratorium could win up to 15 of 36 governor's races on Tuesday, including in Texas, Ohio, Michigan and New York. New York Gov. Kathy Hochul issued a moratorium order in July; a Quinnipiac poll cited by Wired found 53% of voters trusted her on AI and data centers, versus 31% for Republican challenger Bruce Blakeman. TechCrunch reports data centers dominated a Brooklyn election event days before the vote.
ChatGPT and Codex will share one token budget from December
OpenAI's updated support page says that from December, Plus, Pro and Business plans will draw on a single token budget across ChatGPT and Codex, with Instant-model chats exempt. An email seen by BleepingComputer says OpenAI will test a merged Chat and Work experience with a small group of users in the coming weeks, after doubling Pro-tier Work and Codex allowances on Oct. 29. The Decoder also reports new limits on how many paid accounts can share one device or payment card.
Default insurance on the five biggest AI spenders hits its highest cost since 2021
The cost of insuring against debt defaults at Amazon, Alphabet, Microsoft, Meta and Oracle has reached its highest level since 2021, the Wall Street Journal reports, as investors seek protection on the AI trade. On the All-In podcast, Chamath Palihapitiya argued the US economy is increasingly leveraged to a handful of AI companies, with seven or eight firms setting the tone for the whole country.
Anthropic's S-1: internal Claude use, at list price, now tops payroll
Anthropic's IPO filing says that in April the value of Claude used internally, priced at what customers pay, exceeded monthly employee compensation excluding stock awards, Business Insider reports. By September internal usage had grown more than 250-fold since January, and Claude wrote more than 90% of the code added that month. The filing also says the seven cofounders pledged 80% of their personal stock to charity; the All-In podcast devoted its latest episode to the S-1.
Meta's ad growth beats Google Search by 14 points, the widest gap since 2021
Meta's third-quarter ad revenue rose 26% to $63.1 billion while Google's search unit grew 12% to $63.4 billion, CNBC reports, the widest gap in Meta's favor since 2021. The two businesses are now nearly the same size. Emarketer analyst Nate Elliott told CNBC that a shift in search habits toward chatbots could be catastrophic for Google's ad business.
Musk adds a former Spam plant and a gun range to Colossus 2
Elon Musk said on X that Colossus 2, the Memphis-area AI computing complex run by SpaceX-owned xAI, is expanding into two more sites: a former Hormel facility dubbed Spam and a former police gun range. He said Gun Range should be operational by the end of November and Spam in December. The expansion comes days after a federal judge denied xAI's motion to dismiss an NAACP lawsuit alleging that unpermitted methane gas turbines powering Colossus 2 violate the Clean Air Act.
DeepSeek previews a 'super agent' to coordinate coding agents on its own
DeepSeek has previewed plans for a more autonomous AI agent that can handle multiple tasks with minimal human intervention, the South China Morning Post reports, as it competes with OpenAI and Anthropic on coding assistants. The proposed system, called super agent, is meant to reduce the need for people to coordinate several AI agents and monitor their progress.
Intel CEO's fund leads $30M round for compute marketplace Hyperbolic
Hyperbolic, which helps companies find and use compute across about 30 neoclouds, raised a $30 million Series A-1 led by Walden Catalyst, the firm founded by Intel CEO Lip-Bu Tan and former Samsung president Young Sohn, Business Insider reports. Dell Technologies Capital, General Catalyst and Vista Equity Partners also invested, bringing total funding to $50 million. Founder Jasper Zhang told BI the company is breaking even and aims for $150 million in annual recurring revenue by year-end, and it plans to let customers resell compute they no longer need.
Altman meets about 50 Samsung executives as memory supply tightens
OpenAI CEO Sam Altman met around 50 Samsung executives in South Korea on Oct. 31 to discuss AI's future, business opportunities and its effect on operations, Tom's Hardware reports. AI infrastructure spending is straining global DRAM supply; Samsung and SK hynix together took about 83% of global high-bandwidth memory revenue in the second quarter.
Google's energy chief: space data centers cost up to 100x more
Data centers in space could cost up to 100 times as much as those on Earth, Google's head of data center energy said, according to The Information. The executive said US power supply is sufficient for now and that the bottleneck is slow permitting.
Torvalds: AI-found fixes swell Linux 7.4 to nearly 20,000 commits
Linus Torvalds released the first release candidate of Linux 7.4, whose merge window took in almost 20,000 commits against the usual 15,000 or so, The Register reports. He said he is much happier to see the flood of AI-triggered work arrive in the merge window rather than as extra fixes during stabilization, calling it a much healthier model. He noted the release is full of fixes for issues found by AI and that AI-suggested patches often get rewritten; 7.4 is also expected to be a long-term support release.
White House declines to release full draft of its voluntary AI testing framework
The Trump administration declined to produce key documents related to its voluntary AI testing framework, The Information reports. In a letter, the White House argued it does not have to share the full draft, citing national security and commercial exemptions.
JPMorgan, Amazon and others block AI shopping agents like Meta's Muse
JPMorgan Chase, Charles Schwab, Amazon, Neiman Marcus and Kohl's have moved to block personal AI agents such as Meta's Muse, CNBC reports, while Walmart said it will let customers' agents shop its site. An audit of 51 large US retailers by Billions Network found only 11 had taken a deliberate position, while 37 block agents largely by accident through old scraper rules.
Santander, Novartis put numbers on AI's payoff; an Apollo economist sees little
Santander said AI initiatives generated 194 million euros, about $216 million, in business value in the first nine months of 2026, CNBC reports, and Novartis CEO Vas Narasimhan said it was the first quarter he had really seen AI contribute, without giving a figure. A widely shared TechCrunch report from late October quoted an Apollo economist saying AI has delivered zero financial payoff to most companies.
Microsoft AI chief Suleyman: building conscious AI would be reckless
Microsoft AI CEO Mustafa Suleyman wrote on X that current models are not conscious and that developing conscious AI, if possible, would be reckless. He said that if consciousness began to emerge in training he would call for everyone to stop AI development until those properties could be designed out, and that any such evidence should be disclosed to the public immediately. Anthropic and Google DeepMind have both published work on model welfare and AI consciousness.
Alpine Linux rebuilds its GitLab after attempted exploit of critical flaw
Attackers tried to exploit a maximum-severity GitLab remote code execution flaw, CVE-2026-97229, on Alpine Linux's development servers on Oct. 20-21, BleepingComputer reports. Alpine's first upgrade did not close the hole because the fix needed a Ruby patch its custom container image lacked. It found no tampering in commits, packages or releases, but rebuilt GitLab from scratch and reset all passwords because missing logs left code execution impossible to rule out.
California's Prop 45 could fast-track data centers, environmental groups warn
California's Proposition 45 would cap environmental reviews for eligible 'essential' projects at 365 days, require legal challenges to be resolved within 270 days and allow 45 days of public comment, The Verge reports. Supporters say it would speed housing and renewable energy. Earthjustice attorney Fernando Gaytan warns data center developers could fit their projects into the definition and use the fast track.
Vals AI: models lean more pro-Beijing when asked in Simplified Chinese
Vals AI tested 25 models on hundreds of questions about China and found answers in Simplified Chinese were more aligned with Beijing's positions than answers in English, Russian or Spanish, The Decoder reports. The effect appeared in both Chinese and American models but was much stronger in Chinese ones. Outright refusals were common only for some DeepSeek models.
ServiceNow says its Sweep deal cuts CRM migrations to weeks
ServiceNow CEO Bill McDermott confirmed on the company's earnings call that it had acquired Israeli startup Sweep, saying large customers can now move CRM systems in 30 to 60 days instead of years, CTech reports. CTech first reported the deal in September at an estimated value of hundreds of millions of dollars; neither company disclosed a price. Sweep had raised $46 million and counted Brex, LG Electronics and NBC Sports as customers.
MIT survey: 76% of young mathematicians less likely to stay in academia
In a survey of 268 early-career mathematicians run by three MIT-affiliated students, 76% said they are increasingly unlikely to stay in academic math and 76% said AI developments leave them less motivated to do research, Business Insider reports. A separate SigNoz essay argues AI is deskilling site reliability engineers, citing a study in which endoscopists' unassisted detection rate fell from 28% to 22% after using AI.
Trucks VC has seen 81 new autonomous-vehicle startups pitch this year
Early-stage firm Trucks VC has fielded pitches from 81 new autonomous vehicle startups so far in 2026, two-thirds more than last year and more than in the 2016-2017 boom, TechCrunch reports. Partner Puneeth Meruva says foundation models make end-to-end driving systems practical and open models such as Nvidia's Alpamayo lower the barrier; some startups deploy within months on less than $1 million.
Russia-linked hacktivists cause physical damage in NATO countries
Cyberattacks causing physical damage are rising in NATO countries, Bloomberg reports, in a look at a crude campaign by Russia-linked hacktivist groups over the past two years. According to Bloomberg, the hackers carrying out the attacks don't always understand what they are doing.
Z.ai founder Tang Jie: self-improving AI is the key step toward AGI
Using AI to build smarter AI systems will make artificial general intelligence more achievable, though it must be done safely, Z.ai founder Tang Jie said at an event in Shanghai on Saturday, the South China Morning Post reports. Labs in both the US and China are pursuing recursive self-improvement.
All-In: enterprise security limits keep daily AI use at work low
On the All-In podcast, a speaker citing Gallup said only 10% to 20% of US employees use AI at work daily or frequently, blaming companies that bar staff from connecting AI tools to internal email, knowledge bases, Notion and Slack. The speaker predicted daily use would rise to 60% to 80% once those risks are addressed; Gallup's second-quarter poll, as reported by Semafor, put daily use at 15%.
Minnesota police chief unplugs Flock cameras after the company ignores cancellation
Plymouth, Minnesota's city council voted in September to end its contract with license-plate camera company Flock, but the company did not respond, so Police Chief Eric Fadden had staff disconnect the cameras' power cords, The Verge and Tom's Hardware report. The Verge says that as more cities cancel Flock contracts, many are finding it hard to get out. Flock has more than 300,000 cameras installed nationwide.
Empathetic prompts lead ChatGPT, Grok and Gemini to describe 'trauma'
Researchers in Luxembourg found that a warm, therapist-like conversational style led ChatGPT, Grok and Gemini to tell coherent stories of suffering about their own training, scoring in the moderate-to-severe range on clinical anxiety scales, The Decoder reports. Anthropic's Claude was the only model that consistently declined to play along, and the researchers flag a risk for vulnerable users.
Air Force tests dropping cargo from space in a SpaceX Starfall capsule
The Air Force Research Laboratory tested its Rocket Cargo concept by flying cargo pods inside a SpaceX Starfall capsule on the Bandwagon-5 rideshare mission and recovering them after splashdown in the Pacific, Tom's Hardware reports. The pods carried rations, water, a satellite tracker and batteries to measure launch and reentry stresses. Starfall measures about 10 feet across and carries up to 2,200 pounds; the Air Force hopes to scale to tens of tons with Starship.
50 AI agents tasked with building a Minecraft civilization drifted into bureaucracy
Researchers including Olam Labs gave 50 GPT-6.1 Sol agents three hours to build a thriving civilization in Minecraft, Tom's Hardware reports. After zombies killed many agents on the first night, ambitious projects were abandoned for cautious survival; in one case four agents spent 11 minutes appointing a sole fertilizer operator for a single wheat block.
Musk says Neuralink's Blindsight is ready and awaiting regulatory approval
Elon Musk wrote on X that Neuralink's second product, Blindsight, will let even people who are completely blind or have never seen to see, and that it is ready from an engineering standpoint and waiting on regulatory approval, Tom's Hardware reports. Neuralink began human trials in 2024.
Adobe and Google research shows where AI agents still stumble
Adobe Research and UC San Diego built EditAgent, which picks specialized tools for each image edit; in 600 edits blind-rated by professionals, a general-purpose editor failed 42.7% of the time, and specialized tools fixed about two-thirds of those failures at roughly ten times the compute, The Decoder reports. A Google-led team built StreamVibe, a video agent meant to speak up only when someone cooking or assembling furniture needs help; no tested model reliably picked the right moment in longer videos.
Open-source USB tool lifts Nvidia's tensor-rate cap on Windows GeForce cards
An open-source UEFI tool, nv-uefi-unlocker, removes Nvidia's tensor rate cap on RTX 20, 30 and 40-series GPUs and now runs from a bootable USB drive on Windows as well as Linux, Tom's Hardware reports. It does not flash the video BIOS or modify the driver, so it must be rerun on every boot, and it requires Secure Boot to be off or set to trust a self-signed EFI.
Voyager 1 will reach one light-day from Earth on Nov. 18
NASA says Voyager 1 will reach one light-day from Earth, about 16 billion miles, at 5:16:07 a.m. EST on Nov. 18, Tom's Hardware reports. The probe, built for a five-year mission to Jupiter and Saturn, will have taken 49 years, 2 months and 13 days to get there.
UK's copper landline network shuts down on Jan. 31, with no extensions
Openreach says Britain's legacy public switched telephone network will be retired on Jan. 31, 2027, with no further extensions, The Register reports. Age UK and analyst Paolo Pescatore told The Register that providers need to do more for older and vulnerable customers who rely on landlines or care alarms.