_o_
______________|______________
/\ | /\
/ \ | / \
'----' | '----'
__|__
'-----'
══════════════════════════════════════════
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. 88 · Wednesday, 14 October 2026
Nice of you to drop in.
Google opened Gemini 4 Argon to paying users and developers, at a price that from December matches what Anthropic charges for Opus 5.5 and Meta for Muse Wonder. Reuters reports that OpenAI and Anthropic have sought significant control over what the outside safety evaluators they invited in may publish. And US prosecutors charged an OpenAI engineer with hiding a recording device behind a television in an executive conference room.
░░▒▒▓▓ TOP 5 ▓▓▒▒░░
Google opened Gemini 4 Argon to paying subscribers and developers, two weeks after limiting it to vetted cyber defenders (update)
Since Tuesday, Google has been rolling Argon out in the Gemini app to subscribers to Google AI Pro and AI Ultra, and developers can reach it through the API, 9to5Google and SiliconANGLE report; inside Antigravity, Google's coding tool, it requires the Ultra tier. As we covered on October 1, Google first released its new frontier model only to vetted cyber defenders. API access costs $2 per million input tokens and $10 per million output until November 30, then doubles to $4 and $20. That second price is the one Meta set for Muse Wonder on Monday and the one Anthropic charges for Opus 5.5, so three rival labs will soon sell these models off the same price list (9to5Google · SiliconANGLE · Google DeepMind · Meta · TechCrunch · Anthropic).
OpenAI and Anthropic have sought significant control over what outside safety evaluators can publish, Reuters reports
Both companies publicly back the evaluators' independence: Dario Amodei, Anthropic's chief executive, has said they should be able to publish key findings without the company's editorial control, and OpenAI has said they should have editorial independence. In private negotiations, both labs worried that findings could expose trade secrets, two people familiar with the talks told Reuters. Measures under discussion include having lab employees observe the embedded evaluation teams and barring an evaluator placed at one lab from working at another at the same time. The trouble, a person with knowledge of past dealings with the labs told Reuters, is that what a lab calls crucial intellectual property can overlap with information relevant to judging how risky a model is. Anthropic told Reuters that seeking significant control does not accurately characterize all of its work with independent evaluators, though for new evaluator firms it reserves the right to comment on and approve reports; one person familiar with the matter said its talks with some evaluators appeared to have evolved by October 8. OpenAI said it is not seeking approval rights over the full text of evaluators' reports (Reuters via The Star · CTV News).
US prosecutors charged an OpenAI engineer with hiding a recording device in an executive conference room
Abdulrahman Alfozan, a 35-year-old Saudi national who had worked at OpenAI as an engineer since July 2025, was arrested on October 9 and charged with trying to intercept oral communications, the Justice Department said on Tuesday. According to the complaint, staff found the recorder on September 30, plugged into a USB port on the back of a television and wrapped in toilet paper, in a room used for confidential talks about research, products and deals. OpenAI says it used its own AI models to trace the device and identify him, and that it has evidence he meant to plant more (SF Standard · Reuters via Spokesman-Review · The Hacker News).
ASML, the Dutch maker of chipmaking machines, reported €11.7 billion in third-quarter sales and said some customers already want to order for 2029 and 2030
Net income reached €5.0 billion, up from €2.9 billion in the second quarter, and the gross margin rose to 57.5%, the company said this morning. ASML expects fourth-quarter sales of €14.2 billion to €15.2 billion and €44 billion to €45 billion for the full year. Christophe Fouquet, the chief executive, said customers keep raising their equipment forecasts on expectations about the scale and duration of AI investment. ASML's own supply plans for 2027 and 2028, he said, would not fully meet that demand (ASML · Investing.com).
Character.AI, the chatbot startup sued over harm to teenagers, filed for bankruptcy, and Disney opened the bidding for some of its assets at $20 million
Character Technologies, the company behind Character.AI, filed for Chapter 11, listing estimated assets and liabilities of $50 million to $100 million, NBC News reports. Disney's bid, for a nonexclusive license to some of its intellectual property, opens a court-run auction, and the company says it will shut the platform down if it cannot find a buyer. The company has faced a wave of lawsuits claiming its app contributed to mental-health problems and suicides among teenagers. In September 2025 Disney sent Character.AI a cease-and-desist letter over its use of Disney characters; this time Disney is the one asking for a license (CNN · NBC News · Variety · Giga Law).
░░▒▒▓▓ STATS OF THE DAY ▓▓▒▒░░
░░▒▒▓▓ EARNINGS ▓▓▒▒░░
The chip-equipment season opened with demand outrunning supply; ASML's numbers are in TOP 5 above.
The test tomorrow is whether TSMC, which makes the chips, sounds as stretched as ASML, which makes the machines.
░░▒▒▓▓ AI ▓▓▒▒░░
Anthropic shows institutional investors its third-quarter financials today, the first time it has disclosed them. The company is preparing for what could be the largest IPO on record, at a valuation of around $2 trillion, CNBC reported, citing people familiar with the plans. Anthropic filed confidentially with the SEC earlier this year and has not made its S-1 financials public; Reuters reported earlier that it will list in November, though people familiar with the plans could not confirm a date. Anthropic declined to comment (CNBC · Quartz).
JPMorgan Chase is blocking outside AI agents such as Meta's Muse agent and Spear Street Technology's Instinct from directly accessing customer accounts. Jamie Dimon, the chief executive, told analysts on Tuesday that letting just any outside agent in would be too risky for clients and for the bank, and warned of rogue agents that could number in the thousands. JPMorgan uses agents internally, and the ones it plans for customers will be its own. On the same call, the bank reported $15.3 billion of quarterly net income, though even excluding securities losses, earnings of $5.75 a share missed the $5.90 analysts expected (CNBC · Forbes · CNBC earnings).
OpenAI is scrambling to fix safety problems before it ships its next flagship model, The Information reports. According to the report, OpenAI is dealing with fallout from its use of a design known as loop transformers, or recurrent depth, which can make it harder to monitor how a model reasons and could erode a key tool that OpenAI and other companies rely on to catch misbehaving models. The Information named the method in September, as we reported then, and now says Microsoft last week accidentally confirmed that GPT-6.1 Sol uses it, in a blog post it has since edited (The Information).
░░▒▒▓▓ TECH ▓▓▒▒░░
Apple unveiled HomeView, a $299 seven-inch smart-home display, at Tuesday's event in New York (update). The hub we flagged yesterday recognizes up to six household members and showcases Apple's revamped Siri, The New York Times reports, as Apple tries to prove it can still innovate. Preorders opened on Tuesday, and it becomes available on November 13, though not in the European Union, where Apple says Siri AI is currently unavailable (The New York Times · Apple).
Hackers set the refrigerators at around 400 French supermarkets defrosting nonstop overnight into Monday. The attack hit refrigeration controllers running older versions of control software from Danfoss, the Danish equipment maker; some systems were reportedly set to 50°C, leaving food in the heat for several hours. Two supermarkets in Alsace reportedly lost €270,000 of stock, and France's cyber agency, ANSSI, and prosecutors are investigating. Intermarché, Carrefour and E.Leclerc were among the chains hit (BleepingComputer · Cybernews).
░░▒▒▓▓ INFRA ▓▓▒▒░░
A third Ukrainian drone strike shut down Yandex's data center in Vladimir, and the charts hailed as proof that Russian bots had gone quiet do not necessarily track social media (update). Yandex said early Sunday that the site's operation was "completely suspended"; we covered the first two strikes, on Sasovo and Kaluga, on Monday. Cloudflare, whose Radar service tracks internet traffic, said traffic from Yandex's network fell 99% from the prior week early on October 11 and was still down about 89% on October 13. Its bot-traffic chart spread online as evidence that Russian troll farms had been silenced, and President Zelensky said on Monday there were "fewer provocations because there were fewer bots." Cloudflare says "bot traffic" means automated requests "such as crawlers and scripts, not necessarily activity on social media platforms," and Antibot4navalny, which tracks state-sponsored troll farms, said the chart had nothing to do with social media (Yandex Cloud · Cloudflare, Oct 12 · Cloudflare, Oct 13 · Euronews · Cybernews).
NTT Data, a subsidiary of the Japanese telecom group NTT, is in final talks to build Anthropic's first dedicated data center in Japan, a site in Osaka prefecture that could draw up to 500 megawatts, Nikkei reports. That much power would make it one of Japan's largest data centers; construction could cost up to 1 trillion yen, about $6.3 billion, with partial operations as early as 2028. Drawing 500 megawatts from the grid would take time, so the partners will first look at generating power on site, including with fuel cells. Both companies declined to comment (Nikkei Asia · The Next Web).
░░▒▒▓▓ BUSINESS ▓▓▒▒░░
Bumble, the dating app, agreed to be taken private for $1.3 billion by Ithaca Holdings, a new firm that buys companies to remake them with AI. Shareholders get $6.05 a share in cash, a 138% premium to the October 9 close, according to Bumble's filing. Ithaca was founded by Ravi Gupta, a Sequoia partner, and John Hegeman, Meta's former chief revenue officer; the two will become Bumble's co-CEOs. The firm describes its mission as applying frontier technology to renew businesses in a world transformed by AI (Bumble · Yahoo Finance).
Broadcom's chief executive, Hock Tan, said on Monday that many AI investors have no idea what they are buying and behave as if they were in a casino. Speaking in Singapore, Tan took aim at money flowing into AI labs that still lose money training models. His own company finances an AI lab too: Broadcom previously agreed to lend Anthropic up to $42 billion to rent computing capacity on TPUs, the AI chips Broadcom helps Google design (Bloomberg · Tradingpedia · Reuters via The Star).
░░▒▒▓▓ GEOPOLITICS ▓▓▒▒░░
Pope Leo XIV has lung cancer, the Vatican said after surgery on Tuesday. Doctors at Rome's Gemelli hospital removed the lower lobe of his right lung, and tissue tests confirmed adenocarcinoma, a common form of lung cancer; they described the operation as successful and without complications. The 71-year-old pope is expected to stay in the hospital for about a week, and his audiences are canceled through October 20 (Vatican News · CNBC).
Former Congolese president Joseph Kabila launched a political and military front that includes the M23 rebels. The Front for the Refoundation of the Nation, announced on Monday with Kabila at its head and rebel leaders as deputies, says it will end what it calls tyranny by any means necessary, including force. The move complicates a peace process with the rebels that has had limited success; President Félix Tshisekedi's government gave no immediate reaction (AP via La Nación · Reuters via Internazionale).
░░▒▒▓▓ REPORTS ▓▓▒▒░░
International Monetary Fund — World Economic Outlook, October 2026. The IMF raised its forecast for global growth this year to 3.1%, still a slowdown from 3.5% in 2025, before a recovery to 3.4% in 2027, with the AI investment boom offsetting part of the Iran war's energy shock. Inflation is expected to rise to 4.7% this year from 4.1% in 2025, mainly because of higher energy and food prices. Just 32 countries now account for 95% of global growth, against 54 as of April, as growth is reallocated according to each economy's exposure to the war's supply shock and to AI demand (IMF · Fortune · The Star).
░░▒▒▓▓ COMMS DESK ▓▓▒▒░░
Three of the largest US book publishers are quietly using AI for publicity copy, cover art and emails to literary agents, WIRED reported on October 9, citing staff. Employees at HarperCollins, Simon & Schuster and Hachette described chatbots such as Claude and ChatGPT drafting back-cover text without authors' consent; HarperCollins named dozens of staff as AI champions to find uses. Hachette told WIRED it supports AI for operational purposes but not for what it calls creative uses, including communicating with authors and partners (WIRED).
░░▒▒▓▓ ONE MORE THING ▓▓▒▒░░
Anthony Sistilli, a software engineer in Brampton, Ontario, built a prototype license-plate reader around a $180 camera, designed to pick out police cars rather than ordinary traffic, and wrote its software with AI coding tools. After he posted videos of the project, two officers came to his door last month, by his account, to ask what he planned to do with the data, warning that mapping police movements could put officers in danger. He says he told them those were the same worries civilians have about license-plate surveillance (Gizmodo · TechSpot).
░░▒▒▓▓ TRACK OF THE DAY ▓▓▒▒░░
Over and out — the Uncertain Futures desk, freshly swept