░░ ░░ ░░ ░░ ██ ██ ██ ░░
░░ ░░ ▓▓ ▓▓ ▓▓ ▓▓ ▓▓ ░░
▒▒ ▒▒ ▒▒ ▒▒ ▒▒ ▒▒ ▒▒ ▒▒
T H E U N C E R T A I N F U T U R E S
all the news that fits the context window
──────────────────────────────────────────
No. 85 · Friday, 9 October 2026Let's start with the good news: the weekend is in sight. OpenAI's annualized revenue is near $50 billion, not the $70 billion reported last month and repeated here, the Financial Times reports. Samsung expects about $80 billion in quarterly operating profit, the first time any tech company in the world has passed 100 trillion won, according to Reuters. And Washington suspended Microsoft from a green-card program on the day Microsoft's chief executive collected a national technology medal.
░░▒▒▓▓ TOP 5 ▓▓▒▒░░OpenAI's annualized revenue is near $50 billion, not the $70 billion reported last month, the Financial Times reports
OpenAI told investors its run rate was approaching $50 billion as of September, according to the FT report on Thursday. Axios reported the $70 billion figure on September 29, and we repeated it the next day. The FT says OpenAI's own investors produced the $70 billion figure to compare it with Anthropic's, which counts sales made through cloud partners such as AWS and Google Cloud; OpenAI's count does not. The Nasdaq Composite fell 1.25% on Thursday, its worst day since mid-August, with Nvidia, Oracle and Intel among the decliners, CNN reported (TechCrunch · Investing.com · CNN via Yahoo Finance · Reuters via The Star).
Samsung expects about $80 billion of third-quarter operating profit, the first quarterly operating profit above 100 trillion won at any tech company in the world
Samsung Electronics' preliminary estimate on Thursday put operating profit at about 107.4 trillion won, or $80.17 billion by Quartz's conversion, which Reuters called a world first for a technology company. That is nearly nine times the 12.17 trillion won of a year earlier, on revenue of about 195 trillion won, as AI servers soaked up its memory chips. Analysts polled by LSEG had expected 106.1 trillion won, Quartz reported. TSMC, the world's largest contract chipmaker, reported record third-quarter revenue of NT$1.49 trillion ($46.71 billion) the same day, up 50% from a year earlier and above both analysts' forecasts and its own guidance (Quartz via Yahoo Finance · Reuters via Investing.com).
Washington suspended Microsoft, Adobe and six IT services firms from a green-card program, and Microsoft's chief executive collected a national technology medal the same day
The Labor Department will no longer process permanent labor certifications, usually the first step when an employer sponsors a foreign worker for a green card, for Microsoft, Adobe, Capgemini, Cognizant, HCL, Infosys, Tata and Wipro, the administration said on Thursday, accusing the companies of fraud. Vice President JD Vance said the message to Microsoft was that it is "a great American company" but has "got to hire great American workers." Nearly three-quarters of approved H-1B visas go to workers from India, according to The Associated Press. Microsoft answered that 80% of its roughly 6,000 H-1B visa applications last fiscal year were for people it already employed. Right after the suspension, President Trump presented Microsoft's chief executive, Satya Nadella, with the National Medal of Technology and Innovation (TechCrunch · TechCrunch).
Anthropic will drop its blanket ban on personalized vote and campaign targeting from Claude's rules on November 12, nine days after the US midterms
Anthropic said on Thursday that the old ban also caught legitimate civic work, such as nonprofits writing voter information in other languages or election officials sending ballot cure notices, which tell voters how to fix a flawed mail ballot. Voter targeting that relies on deception or misused personal data stays banned under other sections, it said, and the rewritten elections section focuses on barring the use of Claude to deceive voters or disrupt elections. Anthropic says the update reflects that Claude has taken on longer, more independent work, and that most of the changes are meant to clarify existing rules, among them those on fake accounts and influence campaigns, on the software that makes weapons work, on tracking people without consent, and on keeping a qualified human in the loop for uses that can affect someone's health, legal rights or finances. The new policy also adds a ban on "sustained and needless abusive or cruel behavior toward our models", which Anthropic says is meant only for extreme cases where users repeatedly act cruelly with no discernible purpose, not for ordinary frustration, pushback or model testing (Anthropic).
SpaceX agreed to buy nationwide low-band airwaves for its satellite phone service, and shares of AT&T, Verizon and T-Mobile fell about 6% after hours
SpaceX is buying the 800 MHz spectrum portfolio of the private-equity firm Grain Management, a step toward taking on the big carriers directly, Reuters reported on Thursday; Starlink Mobile will pair the new low-band frequencies with the mid-band airwaves it already uses to reach phones from orbit. Terms were not disclosed, and the deal needs approval from the Federal Communications Commission, which has also authorized 15,000 more Starlink satellites. A Verizon spokesman said: "It's just empty airwaves" (Reuters via Yahoo Finance).
░░▒▒▓▓ STATS OF THE DAY ▓▓▒▒░░░░▒▒▓▓ EARNINGS ▓▓▒▒░░Results season reaches the chipmaking supply chain next week, after a Thursday that brought Samsung's record quarter and the Nasdaq's worst day since mid-August.
The test is whether TSMC's outlook matches the demand behind its record revenue.
░░▒▒▓▓ AI ▓▓▒▒░░The three safety researchers OpenAI fired say its misconduct claims are wrong (update). Jasmine Wang, Tomek Korbak and Mikita Balesni, whose dismissal we covered on October 2, deny involvement in a leak to a news outlet about OpenAI model designs that make the models' step-by-step reasoning harder to monitor, and deny dealing with outsiders beyond their jobs. In a letter to OpenAI's safety and security committee and two advisory groups, they say the firings have left former colleagues afraid to speak up in ways that were normal at OpenAI until last week. OpenAI says an investigation found a "pattern of misconduct" that went beyond sharing information with an outside evaluation group (TechCrunch).
Google and Elon Musk's SpaceX will both let their AI agents hand work to Anthropic's Claude. Google's Gemini agent, unveiled at a Google Cloud event on Thursday for business customers first, can be given objectives, not just instructions. It gets its own Workspace account and email address like any co-worker and lets users pick other companies' models, starting with Claude. A day earlier, on October 7, Musk wrote that SpaceX, whose AI division SpaceXAI makes Grok, will use the best back-end model for each task in its Grok Bot, Claude included: "Whatever is most likely to give you the best outcome." Anthropic, a rival to both, will be a supplier to both (TechCrunch · Elon Musk on X).
CrowdStrike found a résumé request in the exposed AI logs of the suspected hacker behind the South Korean bank breaches (update). The prompt named a person called YY with a Chinese university and a location in Guangdong province, details CrowdStrike says likely belong to the attacker, though it cannot definitively link them, The Register reported on Thursday. The logs show the attacker running ARTEX, an open-source Chinese tool for penetration testing, alongside Anthropic's Claude Code, with models including DeepSeek and Z.ai's GLM. At least five lenders were hit in the breaches we covered yesterday, Shinhan Bank among them with about 25,000 customers affected, and lawmakers have approved plans to summon the heads of five major commercial banks to a parliamentary audit on October 19 (The Register · CrowdStrike).
░░▒▒▓▓ TECH ▓▓▒▒░░Margaret Hamilton, who led the team that wrote the Apollo flight software, has died at 90. She died on September 30, MIT said. By 1968 she ran the Command and Service Module software team at MIT's Instrumentation Lab, and more than 400 people were working on Apollo's software; she began using the term software engineering to set the work apart from hardware. During the Apollo 11 descent, a 1202 alarm warned that the computer was overloaded, and her team's software dropped background jobs to keep the landing tasks running, so Houston could let the landing go ahead (MIT News).
░░▒▒▓▓ POLICY ▓▓▒▒░░Britain's data regulator says ten of the biggest AI developers have changed, or promised to change, how they handle personal data. The Information Commission's Office, or ICO, named them on Thursday, from Amazon and Apple to DeepSeek and Stability AI, citing clearer transparency information, stronger ways for people to exercise their rights and tougher assessments of safeguards. SpaceXAI, formerly xAI, is missing because the ICO paused talks with it after opening a formal investigation into Grok. The ICO also opened a six-week call for evidence on AI agents, running to November 20 (ICO).
Singapore told financial firms they remain answerable for the AI they buy from others. Under guidelines the Monetary Authority of Singapore issued on October 7, financial institutions remain accountable for AI used in their services, including AI developed, operated or provided by third parties, and should check that outside systems suit their intended use. The guidelines take effect on October 7, 2027, and firms have until October 7, 2028 to meet the life-cycle controls that include these third-party checks (MAS · MAS guidelines).
The US Treasury disclosed its first penalty under its rules on investing in Chinese technology, and the fine is more than twice the investment. Amidi, the California parent of the startup accelerator Plug and Play, must pay $200,000 because a Chinese fund it controls put about $92,478 into Noematrix, a Shanghai robotics and AI company, in April 2025 without notifying Treasury, the department said on October 7. The penalty itself was issued in July (US Treasury).
░░▒▒▓▓ INFRA ▓▓▒▒░░Broadcom is arranging more than $50 billion of financing so OpenAI can buy the custom chips Broadcom builds with it. Apollo and Blackstone are among the lenders approached for chips in OpenAI's Nexus program, part of a plan to deploy 10 gigawatts of custom accelerators by the end of 2029, The Wall Street Journal reported on October 7; the talks are early. The stock-research site TIKR counts it as Broadcom's third such package since June, after the $60 billion package for Anthropic and others we covered on October 5, and says Broadcom has now pursued financing for both of its biggest AI lab customers. Oracle is separately talking to Apollo and Goldman Sachs about money for a chip deal of its own (TIKR via Yahoo Finance).
░░▒▒▓▓ BUSINESS ▓▓▒▒░░Isomorphic Labs, the AI drug-discovery company, is in early talks to raise money at a valuation of at least $40 billion. The round could value it at as much as $50 billion, Bloomberg reported on Thursday, citing people familiar with the matter, and the financing has not closed. Isomorphic announced a $2.1 billion Series B led by Thrive Capital in May (Bloomberg · Isomorphic Labs).
Nous Research, maker of the open-source Hermes agent, raised $90 million at a $1.5 billion valuation. Robot Ventures led the Series B, announced on October 7, with Nvidia, Union Square Ventures, Menlo Ventures, Samsung and 1789 Capital joining. Nous says Hermes has been cloned, or copied from its public code, more than 24 million times and drives about 2.5% of global AI token usage. The company's annualized revenue reached about $36 million by mid-September, and it expects to pass $100 million by the end of the year, The Wall Street Journal reported (TechCrunch · Nous Research).
Manus, the AI agent startup that Beijing ordered to unwind its sale to Meta, raised more than $500 million. Boyu Capital and IDG Capital led the round for Butterfly Effect, Manus's parent, with Tencent, HSG and ZhenFund joining, TechCrunch reported on Thursday. Meta agreed to buy Manus in December 2025 without disclosing the price, which The Wall Street Journal put at more than $2 billion; Chinese authorities ordered the deal unwound in April, and Manus had to delete some user data as part of the split (TechCrunch · AP via ABC7).
░░▒▒▓▓ GEOPOLITICS ▓▓▒▒░░A Russian glide bomb hit two minibuses in Kramatorsk on Thursday, killing at least 30 people. Another 20 were injured, and strikes on nearby Bilenke killed three more. Donetsk governor Vadym Filashkin ordered public transport halted in Kramatorsk and Sloviansk. US Secretary of State Marco Rubio told Greece's Kathimerini that for now there seems to be no prospect of a negotiated settlement, because Russia has defined victory as taking all of Donetsk, a condition he called unacceptable for Ukraine (Kramatorsk city council · Prosecutor General's Office · Vadym Filashkin · Kathimerini).
France suspended police stun grenades at high-school protests after a 15-year-old lost a hand. Interior Minister Laurent Nuñez announced the suspension on October 7, two days after the teenager was injured in Lens, and said it would last until the facts are clear. More than 6,500 people, most of them minors, have been arrested since students began blockading schools in late September over crumbling buildings and teacher shortages; about 250,000 people marched across France on October 6 (Euronews).
░░▒▒▓▓ ALPS ▓▓▒▒░░Ads for AI jobs in Switzerland rose 32% in a year, with small and medium-sized companies driving most of the growth. Nearly 5,000 AI positions were advertised between July 2025 and June 2026, and small and mid-sized firms accounted for 76% of the increase, according to a study by the Lucerne University of Applied Sciences and Arts (HSLU). The public sector grew fastest, at 67%, led by the canton and city of Zurich and the Federal Office of Public Health. Only 5% of the openings were permanent junior jobs, and entry-level roles often asked for two to three years of experience (swissinfo.ch).
Ecosia, the Berlin search engine, dropped the French lab Mistral as its AI supplier in favor of open models, Chinese ones included. Founder and chief executive Christian Kroll said Mistral had fallen a year behind and suffered regular outages because Ecosia was too large a customer, heise online reported. Kroll also said the high share of nuclear power in France's energy mix clashed with Ecosia's sustainability goals. Ecosia now works with the hosting platform Melious, whose range of open models includes China's Qwen, GLM and Kimi. Kroll told Politico the switch had roughly halved its costs while improving quality (heise online · Politico).
░░▒▒▓▓ ROBOTICS ▓▓▒▒░░Mecka AI, which pays people to record their everyday tasks for robot training, raised $60 million in a Series B led by Sequoia (update). Contributors wear body sensors and use smartphones while doing routine activities, and the motion data trains humanoid and other robots. Nvidia and Microsoft's venture fund M12 joined the round, which Mecka announced on October 7. The announcement gives no valuation; the figure of about $500 million we reported on September 14 came from TechCrunch (Mecka · TechCrunch).
░░▒▒▓▓ PAPERS ▓▓▒▒░░DeepMind Institute — scientists say AI saves them just under seven hours a week, and nearly half spend a good chunk of that checking its work. Alex Imas, Google DeepMind's director of AGI economics, and James Manyika, Google's president of research, labs, technology and society, set out the findings in an essay published on October 8 by the DeepMind Institute (DMI), a platform started by researchers from Google and Google DeepMind. The essay draws on a September paper by Google researchers and academic co-authors. The paper combines 15 million Gemini interactions, an inventory of more than 2,600 specialized AI models and a survey of more than 600 scientists in the US and UK. Nearly half of those surveyed use AI every day, and just below half say their main bottleneck has moved to lab testing and clinical trials. Almost half, especially junior researchers, say AI pushes them toward incremental questions. For comms teams, the same study gives labs a productivity number and critics a line about AI narrowing science (DMI · AI in Science: Early Insights (PDF)).
░░▒▒▓▓ REPORTS ▓▓▒▒░░Air Street Capital — State of AI Report 2026. The ninth annual report from the investor Nathan Benaich, published on Thursday, calls the frontier a three-lab race and puts the capital spending that Amazon, Alphabet, Microsoft and Meta have guided for 2026 at about $733 billion, a total not limited to AI. It also cites Anthropic's internal index, by which Claude led 26% of measured model research and development work in August, up from under 1% in February. In that index, "leads" means Claude completes most of a task while a human supervises, and none of the measured work is fully autonomous (State of AI).
SemiAnalysis — Beijing Will Not Pace the Frontier. The research firm counted 857 model releases from nine leading Chinese developers since 2021 and found a published safety result for 31 of them, 3.6%, and for only nine at or before launch. No Chinese frontier text model has shipped with a dangerous-capability evaluation, a test of what harm a model could help cause, across the risk areas named in international scientists' safety statements; a cyber note on Z.ai's GLM-5.3 came closest. The firm concludes that Beijing's rules govern what AI applications say and do to people, not what frontier models can do (SemiAnalysis).
░░▒▒▓▓ COMMS DESK ▓▓▒▒░░OpenAI says it disrupted two influence operations that hid behind a fake think tank and fake journalists. In a report published on October 8, OpenAI says Russian operators ran a sham think tank, the Social Research Center, apparently through an invented persona named Mia Clark, and staffed it with people in Latin America who, the evidence suggests, did not know they were working for a Russian group. That operation targeted politics in Argentina and Bolivia and Ukraine's reputation. OpenAI rated it Category 5 on a scale of 1 to 6 that measures how far an operation spreads, the first at that level it has disrupted since it began reporting on such operations. Iranian operators used seven fake journalist bylines to place almost 100 articles in about a dozen outlets, and OpenAI banned the accounts behind both. Takeaway: a pitch from a columnist nobody on the desk has met deserves a phone call first (OpenAI).
░░▒▒▓▓ ONE MORE THING ▓▓▒▒░░Amazon keeps a page called About You in its shopping settings that lists what it has concluded about customers from their purchases, and TechCrunch explained on October 6 how to find it. The guide followed a Threads post that passed a million views in under a day: its author had discovered that Amazon believes she "has flat buttocks", a conclusion she traced to a pair of "butt scrunch leggings". Other inferences are tamer, from photography to vinyl records (TechCrunch).
░░▒▒▓▓ TRACK OF THE DAY ▓▓▒▒░░Schönes Wochenende — the Uncertain Futures desk