▒▒▒▒▒▒▒▒
▒░VOTE░▒
████████▒░░░░░░▒████████
▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓
░░░░░░░░░░░░░░░░░░░░░░░░
──────────────────────────────────────────
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. 101 · Monday, 2 November 2026Hello, and welcome to a new week. Apologies for the late-night edition: the news ran long, and the context window did not.
OpenAI has secured less than half of the $30 billion round we flagged on October 23, The Wall Street Journal reports. S&P Global moved Meta's credit outlook to negative, and Intel had its worst day since 1974. Tomorrow, US voters pick governors in 36 states, and in many of those races data centers are on the ballot in all but name.
░░▒▒▓▓ TOP 5 ▓▓▒▒░░OpenAI has secured about $11 billion of the $30 billion it set out to raise and has cut its target to about $20 billion, reports say (update)
As we covered on October 23, OpenAI was raising money at a $1.25 trillion valuation ahead of a possible listing. After tepid interest from some large investors, it had about $11 billion in commitments, The Wall Street Journal reported on Friday, citing people familiar with the matter. The valuation itself sits below the $1.4 trillion OpenAI had been aiming for. The lower target of about $20 billion comes from a Financial Times report cited by PYMNTS, with family offices and wealthy individuals expected to fill much of the gap. None of these figures come from OpenAI, which has not commented on them. The timing is awkward. Anthropic, its closest rival, could go public as early as this month (The Wall Street Journal · PYMNTS · The Decoder).
S&P Global moved Meta's credit outlook to negative, and Blackstone told investors Anthropic may get a junk credit rating after its IPO
S&P acted on Friday, citing uncertainty over Meta's capital spending and cash flow. It now expects Meta's free operating cash flow to run $22 billion negative in 2026 and $71 billion negative in 2027, before a smaller $39 billion deficit in 2028, Fortune reports. Meta says its credit profile remains very strong. An outlook change is not a downgrade, only a signal of which way S&P leans. The same day, the Financial Times reported that the investment firm Blackstone now expects Anthropic to land below investment grade once it lists, reversing its earlier view. A junk rating would raise Anthropic's borrowing costs and shrink the pool of investors allowed to lend to it. Microsoft, whose cash flow is still positive, sold at least $25 billion of bonds on Thursday, its first US bond sale since 2017 (The Next Web · Fortune · Financial Times · The Next Web).
Intel's shares fell 29% on Friday, their worst day since 1974, and pulled the rest of the chip sector down with them (update)
The 23% after-hours slide we covered on Friday deepened once regular trading opened. The Nasdaq Composite slid 2.9% on the day, Qualcomm 11%, Broadcom more than 5% and Nvidia 4%, CNBC reported. Benzinga put the chip sector's combined loss at about $1 trillion in market value. Intel's quarterly results had beaten estimates, but investors sold the stock on its forecast (Benzinga · CNBC · CNBC).
Candidates who back pausing data-center construction are running in 21 of the 36 US governor races on Tuesday and could win up to 15, The Information calculates
In five of the 21, both major-party nominees support a moratorium, by The Information's count of platforms and public statements. SemiAnalysis, the chip and data-center research firm, says most of the 2027 build-out is already under construction, so the fight is over 2028. If every pause candidate won, the states they would govern hold about 30 gigawatts of capacity planned for that year. Texas, which SemiAnalysis expects to add close to 30% of new US capacity in 2028, has frozen environmental permits since September 21. New York's governor, Kathy Hochul, signed the first statewide moratorium. By SemiAnalysis's reading, the dividing line now runs within both parties, between pausing construction and letting developers keep building on stricter terms (The Information · SemiAnalysis · Newsweek).
Google's Project Suncatcher satellite sent back its first AI answer from orbit, and it was a haiku about being in space
Google said on Friday that a Gemma 4 model, running on one of its data-center-class TPU chips aboard the prototype satellite built with Planet Labs that we covered before its launch, answered a prompt as the satellite passed over Svalbard, Norway. Suncatcher tests whether solar-powered data centers could one day run in orbit. Next, Google says, it will check how the chips cope with heat and radiation, and two linked satellites are planned for 2027 (Google · Notebookcheck).
░░▒▒▓▓ STATS OF THE DAY ▓▓▒▒░░░░▒▒▓▓ EARNINGS ▓▓▒▒░░After Friday's chip selloff, Apple, AMD and Arm report into a nervous market this week.
The test is whether any of the three gives investors a forecast to buy rather than sell.
░░▒▒▓▓ AI ▓▓▒▒░░OpenAI, Google, Microsoft and Anthropic will all honor one robots.txt signal that tells them not to train on a website. The four companies said on Friday that they will respect a shared signal, set out in a draft that Anthropic's Kevin Kelley submitted to the Internet Engineering Task Force, so site owners no longer need a separate opt-out for each lab's crawler. Robots.txt is the file that tells crawlers what they may visit, and a single rule in it will eventually cover all four. OpenAI says it aims to implement the signal in the coming months, Microsoft targets February 1, 2027, and Google says 2027. The signal covers training only. It does not block crawling, search indexing or AI answers that quote a page, and it removes nothing already collected. On the same day, OpenAI published a post collecting court briefs that argue that fair use, the legal doctrine it relies on to train on copyrighted work, benefits everyone (OpenAI · OpenAI · The Next Web).
An Accenture employee's personal Claude agent ran on a Pentagon network for two months before anyone caught it, sources told WIRED. WIRED reported on Thursday that the employee is alleged to have connected a secure Defense Department system to a personal Claude account so an AI agent could write and deploy code. Pentagon staff are currently barred from using Claude, WIRED notes. The vulnerability has since been patched, the sources said. Accenture told WIRED the allegations had been independently investigated and found to be without merit, without saying who investigated; the Pentagon said it does not comment on its networks, and Anthropic pointed WIRED to its usage policy (WIRED).
SemiAnalysis launched GDPx, a benchmark of long agent tasks in 11 high-value fields, and says the best model scores 12%. The fields include chip design, cybersecurity, finance, law and recursive self-improvement, the scenario in which AI improves AI, with tasks built by specialist data firms and run for up to eight hours each. Every model scored essentially zero on the finance and legal tasks. SemiAnalysis ranks Anthropic's Opus 5.5 the best all-round model and OpenAI's GPT-6 Astra the best at self-improvement tasks, and says safety filters drag down some top models in cyber and biology. The research firm is building forecasts of OpenAI's and Anthropic's revenue for its subscribers; now it grades their models too. GDPx is SemiAnalysis's own benchmark, published overnight, and no outside group has yet rerun it (SemiAnalysis).
░░▒▒▓▓ TECH ▓▓▒▒░░The FBI blamed North Korea for the $6.3 billion Binance theft, and Binance reopened withdrawals on Sunday, saying it had covered about 95% of the stolen funds with its own money (update). As we covered on Friday, Binance paused withdrawals after the hack. In a Saturday statement, the FBI tied the theft to TraderTraitor, the North Korean group behind last year's $1.5 billion hack of the crypto exchange Bybit, and asked crypto firms to block the addresses used to launder the money. President Trump said that night that it was not yet known whether North Korea was responsible. Binance began reopening withdrawals in phases on Sunday and said it had covered about 95% of the stolen amount, up from 90% in its previous update; customers then pulled a net $3.64 billion off the exchange, CNBC reports (NBC News · PBS NewsHour · CNBC).
░░▒▒▓▓ POLICY ▓▓▒▒░░Canada joined the EU's network of supercomputers on Friday, at the EU–Canada summit in Montreal. It becomes the 39th participating state in EuroHPC, the EU's joint supercomputing program, which has bought 12 supercomputers so far, and Canada will be able to use its supercomputing, AI and quantum resources. As we noted on Friday, Prime Minister Mark Carney has also said Ottawa would invest up to C$1.6 billion in Cohere, the Canadian AI lab. Evan Solomon, Canada's AI minister, told BetaKit the country would be open to adding its planned national supercomputer to the European network once it is built (EuroHPC · BetaKit).
░░▒▒▓▓ BUSINESS ▓▓▒▒░░Cognition, the maker of the Devin coding agent, is weighing a new round after investors offered valuations of $70 billion to $80 billion, Business Insider reported on Wednesday. Its last round, announced on September 8, valued it at $48 billion. The talks are early, and no round size or lead investor has been reported (Business Insider · The Next Web).
░░▒▒▓▓ GEOPOLITICS ▓▓▒▒░░Bulgaria's incumbent president, Iliana Yotova, won Sunday's runoff with 57.6% of the vote, final results from the Central Election Commission show. Her challenger, former caretaker prime minister Andrey Gyurov, took 39.5% and conceded. Yotova, seen as friendlier to Moscow than her pro-Western rival, becomes Bulgaria's first elected woman president. Bloomberg reads the win as strengthening Prime Minister Rumen Radev's hold on power (Novinite · Bloomberg · Novinite).
Gadi Eisenkot, Netanyahu's challenger and leader of the Yashar party, left the door open on Saturday to a government backed by Arab parties (update). As we covered on Friday, the final count gave Prime Minister Benjamin Netanyahu's bloc 53 seats, Eisenkot's 48 and the Arab parties 19. Eisenkot said any lawmaker who accepts four principles, among them recognizing Israel as a Jewish, democratic, liberal and equal state and Hamas as a terrorist organization, could join the government he is trying to form. He had repeatedly said he did not intend to form a government with the Arab parties. On Sunday, Noam Sohlberg, who chairs the Central Elections Committee, said that even if suspicions of foreign interference are substantiated, they do not undermine the integrity of the vote or its results as a whole (Haaretz · The Jerusalem Post · The Jerusalem Post · The Jerusalem Post).
Hackers used the emergency-alert apps of both Ukraine and Russia to send fake warnings, as a local ceasefire expired (update). Late on Sunday, Ukrainians received false alerts of an imminent nuclear threat, high radiation and a chemical threat, some in error-strewn English, and Russia's Emergency Situations Ministry said hackers used its app to send a false alert across 33 of the country's 89 regions, AP reports. It was not clear whether the attacks were related. The Kremlin's spokesman, Dmitry Peskov, said Russia's state agencies are under constant threat of cyberattacks, a growing number of them using artificial intelligence. As we noted on October 28, a limited ceasefire in Russian-occupied Oleshky let aid in and civilians out; it ended at midnight at the start of Monday, and Ukrainian officials reported Russian drone attacks on the Kherson region hours later (PBS NewsHour).
░░▒▒▓▓ ROBOTICS ▓▓▒▒░░Meta built a humanoid prototype called Max with Fourier, a Shanghai robot maker, in three months, The Information reported on Saturday. The robot weighs roughly 106 pounds, a second version is in the works, and Meta wants it to inform a more polished home robot it could eventually sell, people familiar with the project told The Information. Meta told The Information the work with Fourier is internal research and that it has no plans to sell these units. No launch target has been set. Fourier began in rehabilitation robotics before moving to humanoids (The Information · The Next Web · Fourier).
░░▒▒▓▓ PAPERS ▓▓▒▒░░Epoch AI — a freely downloadable AI model turned Firefox's published security fixes into working attacks before the fixes reached users. Researchers at Epoch, an AI research institute, used GLM 5.3, an open-weight model from China's Z.ai, to build exploits from patches Mozilla had published but not yet shipped. The exploits read local files and worked against six Firefox releases from mid-September until Mozilla's October 27 update, the institute reports. Every working attack also chained a second, unpublished flaw that the model found itself; Mozilla patched it in that update, which Epoch suspects makes such attacks much harder. Epoch concludes that widely available open models can now do much or all of this work themselves. The same setup did not produce a Chrome exploit. For anyone writing a company's security messaging, the window between a public patch and the user's update now counts as exposure (Epoch AI · Unite.AI).
░░▒▒▓▓ REPORTS ▓▓▒▒░░Axios and CivicScience — AI Sentiment Index, first release, published on Friday. Americans' net sentiment toward AI came to 41.8 on a scale of 0 to 100, from online polling of more than 7,300 adults between October 1 and 26. Economic growth was the only one of six measures with a positive score, at +10, while the expectation that AI will create jobs scored −30.4. Axios headlined the result as a jobless AI boom. The index will be released monthly from December (Axios).
░░▒▒▓▓ COMMS DESK ▓▓▒▒░░Publishers have started selling ads that only AI agents see. TIME runs question-and-answer blocks, labeled as sponsored and produced with the ad-tech firm Mobian, on machine-readable versions of its pages, so a brand's message reaches the assistant that does the reading, The Information reported last week. Ally Bank and the Project Management Institute were among the first buyers, and the AI search engine Perplexity blocked the ads from shaping its own answers, calling them deceptive, Digiday reported in August. For communicators, a brand's account of itself now reaches an audience that never sees the page (The Information · Digiday).
░░▒▒▓▓ ONE MORE THING ▓▓▒▒░░David Durman, a research administrator at the University of Kentucky, won a religious accommodation, approved on September 25, that excuses him from personally using AI at work. He is a humanist, and the American Humanist Association calls it the first known case for a humanist in the US. He may still be expected to attend meetings involving AI and submit material through platforms that use it, so his faith covers his own prompts but not his colleagues' (Gizmodo · Newsweek).
░░▒▒▓▓ TRACK OF THE DAY ▓▓▒▒░░Gute Woche — from the Uncertain Futures desk