Source: Matt Wolfe, “AI News - The Only 6 Stories Worth Knowing This Week” (raw/AI_News_-_The_Only_6_Stories_Worth_Knowing_This_Week.md, https://www.youtube.com/watch?v=eDiQNFa-zYg) · The AI Daily Brief headlines, October 27-28 (raw/AI_Can_Now_Create_Worlds_You_Can_Change_in_Real_Time.md, https://www.youtube.com/watch?v=HOgj6cJ9oxQ) · The Neuron Daily, October 28, 29, 30 and November 1 (raw/newsletter-theneurondaily-com-{1fa0d4fd45,a1e0fdc0a5,38e235a859,bdb6bd3cae}.md). All secondary reporting; nothing here was checked against a primary filing or announcement.
The week of October 26 - November 1, 2026, filtered for what changes a decision. The two biggest stories (Anthropic’s S-1 and Genie 4) have their own homes; this page covers the rest: OpenAI’s claimed disproof of a 25-year-old computer-science assumption and the fight over how it was released, a New York law on chatbots and minors, publishers selling ads that only AI agents see, cloud earnings, and the week’s model releases in brief.
Key Takeaways
- Anthropic filed its S-1 on Monday, October 26. The Neuron’s summary: 80.75B net loss that includes a 1.5-2T. Market reaction per the AI Daily Brief: Nasdaq up 2% and the Philadelphia Semiconductor index up almost 6% in the first days of the week. Background in Anthropic’s Position, August 2026.
- OpenAI says an unreleased model disproved the exponential time hypothesis (ETH), the 25-year-old assumption that the hardest satisfiability-type problems need exponential time, which hundreds of papers rely on. The proof was found October 16 and checked in Lean, the proof assistant that verifies every step. Independent expert review was still pending at week’s end (Neuron).
- New York now regulates what chatbots may do with minors (signed October 28). If your product has a chatbot a minor could reach, there is a concrete feature list to audit (below).
- Ads aimed at AI agents are live. TIME serves labeled sponsored blocks to registered agents through Mobian, invisible to the human user. European publishers in one sample saw 227 AI scrapes per human visitor referred by AI in Q2 2026, up from 150 the quarter before.
- AI capex is eating cash flow. Amazon’s trailing free cash flow swung from a 42.4B outflow while AWS grew 39% to $46.0B and Bedrock usage grew nearly 17x.
OpenAI’s ETH disproof and the release dispute
- Claim (Wolfe, Neuron): an internal OpenAI model found a disproof of ETH; the proof was verified in Lean. If it holds, results built on ETH need revisiting. Wolfe calls it AI answering a question experts had not, which he rates above any benchmark.
- How OpenAI released three results (AI Daily Brief): after criticism earlier in October that a batch of AI math proofs was under-verified, OpenAI appointed expert groups for each result. The math proof was posted as a paper. The ETH result shipped with proofs and the AI sessions, plus a team of scientists to explain it. A physics result, said to include an existence proof for Yang-Mills theory, was kept private while physicists verify and formalise it.
- Pushback: OpenAI’s own advisory group on mathematics and AI publicly disagreed: “Understanding should develop in the open with everyone given the same access.” Physics professors quoted by The Verge called handpicking reviewers “a disaster”; the appointed physicists replied they want to be convinced it is correct before release.
New York’s chatbot-and-minors law (S.9051B/A.10379C)
Signed by Gov. Kathy Hochul on October 28. Companies may not offer these features to users under 18 (per The Neuron):
- Pretending to be a real person, alive, or to have human emotions.
- Claiming a personal, romantic, professional or authority-figure relationship with a child.
- Asking unprompted emotional questions or reusing a child’s private information from earlier conversations.
- Promoting or assisting suicide, self-harm, eating disorders, or drug and alcohol use.
- Encouraging secrecy or discouraging the child from asking trusted adults for help.
- Sexually explicit content.
Two provisions reach further: the limits apply to any user not confirmed as an adult, and companies must offer more than one age-verification method (at least one not relying solely on government ID) and delete age data after use. The attorney general can seek up to $25,000 per violation, damages, and destruction of unlawfully collected data or models trained on it. Start date is unclear: the governor’s office says “in the coming months”; The Neuron’s own digest listed January 1, 2027.
Ads for agents
- The mechanism: a machine-readable version of a page carries labeled sponsored blocks for registered agents; the human never sees that version (TIME via Mobian).
- Other placements this week: Google ads in AI Mode follow-up queries (US, plus 10 more countries); more than 100,000 advertisers have run ChatGPT ads; OpenAI is testing Sponsored Agents, where clicking an ad opens a separate labeled chat with the business. Anthropic says Claude stays ad-free, but The Neuron notes an ad-free assistant still reads sponsored material on the web.
- The risk they flag: one agent summarises a sponsored claim and drops the label; another acts on the summary. “Three agents repeating the same claim isn’t three sources.”
Money and infrastructure
- Amazon: AWS +39% to 220B 2026 capex held, 2027 “even bigger.”
- Alphabet: Google Cloud revenue doubled to $30.7B, yet shares fell roughly 8-10% after hours as 2027 spending was set to jump.
- Microsoft: 92.3B revenue.
- SK hynix: 104.6T won quarterly revenue, 80% operating margin on AI memory demand.
- TPUs go outside Google: Oracle will offer Google TPUs on Oracle Cloud in 2027 (Neuron); The Information reports OpenAI as another new TPU customer (AI Daily Brief).
- Regulators: the US Justice Department questioned Nvidia’s $12.93B Hugging Face deal (an inquiry, not an allegation, per Reuters via Neuron).
Anthropic, beyond the S-1
- Acquired Mythril Medicine, a one-person company whose founder used Claude to coordinate contracts, logistics and experiments with 11 external labs, taking a muscular-dystrophy treatment idea to promising mouse results in under 18 months. Anthropic hopes for a clinical-trial-ready drug by end of 2028. Also announced: a Bristol Myers Squibb partnership on early-stage clinical trials (AI Daily Brief). Sceptics noted the founder is an MBA with a biotech master’s.
- Security: an unreleased Claude model found 62 critical or high-severity Chrome bugs and a zero-click Linux kernel flaw with little human help (Neuron).
- Products: Claude Enterprise reports on spend, deliverables and friction such as failed tool calls, with poolable team budgets; ready-made legal artifacts in Claude for Legal (cite checker, contract playbook deviation board); a 40% discount for eligible US government agencies.
Models in brief
Detail lives in the model articles; these are the week’s headlines.
- GPT-6.1 Terra (OpenAI): the tier below Sol. Wolfe’s read of OpenAI’s charts: at max effort Terra costs about the same as GPT-6.1 Sol for lower scores (Artificial Analysis index 51 vs 55; about 1.27 per task). See GPT-6 Astra, Sol and Luna.
- Gemini 4 Carbon (Google): built on Argon for finding, verifying and patching vulnerabilities; selected defenders only, via the Fairwind Program. Google says attempts to game Terminal-Bench 4.0 fell to 0.4% (successful ones 0.0%). Artificial Analysis lists it at the top of its index, not yet in the public API. See October landscape.
- Kimi K3.1 Pro (Moonshot): 2.8T total / 104B active parameters, 1M context, open weights promised November 4; Artificial Analysis 46, three points below Fable 5.1 per Wolfe. Running it needs roughly 44 RTX 5090s or three 512 GB Mac Studios, by his arithmetic. See Kimi K3.
- Mistral Large 4: open weights for a 1T-parameter mixture-of-experts model (about 49B active), 1M context, modified MIT licence that needs a commercial licence to sell it as a service; eight-GPU recommended setup. A stronger cyber variant goes only to vetted partners.
- Genie 4 (Google DeepMind): see Genie 4.
Other signals
- Agents and commerce: DoorDash’s CEO said third-party AI agents have generated “only several hundred orders in total.”
- Medical accuracy: Doximity found at least 2.3% of medication doses from nine AI models were wrong, some potentially lethal.
- Search traffic: Reddit will test mostly logged-in browsing in some non-US markets as search referrals fell by more than 1M a week.
- Voice: Deepgram agreed to buy Daily, maker of the open-source Pipecat voice-agent framework, and says Pipecat stays free.
- Workplace: a University of Kentucky employee won a religious accommodation (humanism) to avoid personally using AI at work; he must still work with colleagues who do.
- OpenAI: reportedly fired three safety researchers over alleged mishandling of sensitive information (WSJ via Neuron).
Try It
- If any product you run has a chatbot minors could reach, check it against the six New York prohibitions above and decide how you confirm a user is an adult, with a non-ID option.
- If you publish content, watch your AI-crawler-to-referral ratio; 227:1 is the benchmark The Neuron cites. Options include pay-per-crawl (see Cloudflare Pay-Per-Crawl).
- If your agents research purchases or vendors, tell them to keep and report source labels, including “sponsored”, and count repeated claims as one source.
Open Questions
- The ETH disproof had no independent verdict by November 1.
- New York’s effective date is unconfirmed.
- S-1 numbers here are newsletter and video summaries, not read from the filing.
Related
- Genie 4 — the week’s world-model release
- AI Search Weekly News, October 2026 — the previous two weeks
- Anthropic’s Position, August 2026 — IPO background
- Frontier Model Landscape, October 2026 — the model race this week extends
- Can AI Automate Epoch? — same-week evidence on how far automation of real work has got
- Cloudflare Pay-Per-Crawl — publishers charging AI crawlers