Daily Brief Nov 10, 2026 · ← Back to this issue
How we checked every item today except "Also happened"
For readers who want to dig in. For what happened, in brief, go back to this issue.
This issue draws on material from November 9 to early November 10. Last night's collection took in 860 pieces, 311 of them read in full, and this issue uses 26 outside sources with checkable links; the later sections also draw on material from the past week. The full edition of this issue is the archive of record.
1. [Today] Anthropic releases Fable 5.2 and Mythos 5.2, and in its own FrontierCode 2.0 run Fable 5.2 matches Opus 5.5's best score at about half the cost per task
Not proven yet： the FrontierCode scores and cost ratio are Anthropic's own; no outside rerun is attached
How we checked
The main source is the text of Anthropic's release page. The full methods for each benchmark are in a separate system card Anthropic published, not on the release page.
2. [Today] Anthropic files a second amended IPO registration, and an exhibit commits Google to buy US$10 billion of Anthropic stock at the offering price
Not proven yet： US$78 is only a fee-calculation ceiling, not the IPO price
How we checked
The main sources are the fee table and the EX-10.15 and EX-10.16 exhibits Anthropic filed on the SEC's EDGAR system. That Google is a TPU supplier comes from the October 26 registration, which SecondSource cited on November 2.
3. [Today] According to the New York Times, the Commerce Department's Megaspeed probe went to the Justice Department after NVIDIA's general counsel contacted investigators
Not proven yet： both outlets' reports rest on anonymous people familiar with the matter
How we checked
The main sources are reports from two newsrooms, the New York Times and Bloomberg; this item cites posts on X by Times reporter Ana Swanson and Bloomberg reporter Mackenzie Hawkins. Investing.com relays the Times and doesn't count as a third outlet. Both outlets rely on anonymous people familiar with the matter. NVIDIA's and the Commerce Department's responses come through the reporters. The Times reporter says Bloomberg confirmed the paper's reporting; which points Bloomberg confirmed isn't in either reporter's posts. Singapore's detention of the executive appears only in Bloomberg's account of what Teter asked Kessler. Singapore's Ministry of Home Affairs said on October 31 that the executive was not personally under investigation. How that statement fits the detention account isn't clear yet (SecondSource, November 9).
4. [Today] OpenAI describes Control Room, its central stop system for research models, and says output stops within 35 seconds at most, by OpenAI's own figure
Not proven yet： the 35 seconds is OpenAI's own ceiling, with no latency distribution or test method
How we checked
The main sources are OpenAI's blog post and the incident report on OpenAI's alignment blog, both published by OpenAI. Guarding against inference engines started without a Control Room connection relies on scanning; the post doesn't give the scan frequency or the miss rate.
[Today] Delta Electronics, a maker of power and cooling components, reported October consolidated revenue of NT$74.666B, up 30.1% from a year earlier, and the announcement gives no product breakdown.
How we checked
[Today] Wharton professor Ethan Mollick, after trying Fable 5.2, says the model is "almost" artificial general intelligence, and that what it lacks is judging whether its work is good enough.
How we checked
[Today] Venture capitalist Tomasz Tunguz estimates the US will have to buy AI model output from data centers abroad as early as 2028, with the timing set by how many planned data centers can't get connected to power.
How we checked
This is one analyst's estimate. Goldman Sachs estimates US data-center power capacity rising from 41 GW in 2026 to 66 GW in 2027. Tokens produced per watt grow about 2.9 times a year, a rate Tunguz combined from three new chips' gains over their previous generations; the post doesn't say how they were combined.
[Today] An Anthropic research post says Fable 5.2 produced three manufacturable circuit-board prototypes from one prompt, and scored 69% on partner Diode's internal layout grading, against 25% for Fable 5.
How we checked
The research post doesn't give the number of test items or the scoring method.
[Looking back] (papers from 2023 to 2025) Three research groups found that reward models and AI judges favor longer answers.
How we checked
According to the authors' self-reported experiments, preference scores and model rankings produced by AI judges lean toward longer answers unless answer length is controlled. Chatbot Arena is a public leaderboard built from users' votes between two anonymous models' answers. These experiments tested only the judge models of their time, and no one has directly measured the effect on leaderboard rankings.
[Today] Google Cloud lets Kubernetes join TPU capacity into larger slices or split it into smaller ones at scheduling time, and Google says startup is up to five times faster.
How we checked
[Today] OpenAI listed ChatGPT Enterprise on AWS Marketplace, so companies can count eligible spending toward their AWS spending commitments, the usage they committed to AWS in advance; the conditions aren't public.
How we checked
[Looking back] (SecondSource reading of July 14) Spending on model inference will split two ways: open-ended problems pay for thinking time, while narrow, high-frequency work moves to distilled small models.
How we checked
Still missing today. The gap that most affects judgment is the price range and financial figures in the main body of Anthropic's IPO registration; this issue's evidence goes only as far as the fee table and two exhibits. The exact date the Megaspeed case moved to the Justice Department, and Megaspeed's own response, aren't in the two reporters' summaries. The full methods for Fable 5.2's benchmarks are in the system card, not on the release page. Delta's monthly revenue has no product breakdown.
Sources & accounting
The past 24 hours. 860 new pieces came in from November 9 to November 10: 568 academic papers, 161 social-platform posts, 77 company and personal blog posts, 13 from a second academic-paper feed (counted separately from the academic papers), 12 subscription newsletters, 12 linked articles, 11 company filings and 6 show transcripts. Of those 860, 311 were read in full, 190 were screened out as off-topic (10 of them before fetching), 6 were only indexed without a full read, 1 was held for another try, and 352 haven't been read yet. Academic papers, 568: 213 read, 77 screened out (1 of them before fetching), 278 not yet read. Social-platform posts, 161: 47 read, 80 screened out, 2 only indexed, 1 held for another try, 31 not yet read. Blogs, 77: 28 read, 16 screened out, 33 not yet read. Second paper feed, 13: 10 screened out (9 of them before fetching), 3 not yet read. Newsletters, 12: 8 read, 4 screened out. Linked articles, 12: 9 read, 3 screened out. Company filings, 11, of which 9 are Anthropic's IPO filings: 4 only indexed, 7 not yet read. The 6 show transcripts were all read. None of them came from last night's three podcast feeds, which found no new episodes.
One-time catch-up. No new one-time catch-up material came in this window. The catch-up material already in the database is dated between July 1 and November 9 this year, counted separately from the past 24 hours above, and none of it was used in this issue.
Source concentration. Main-line items 1 and 2 and Research & models item 1 all come from Anthropic's own releases or filings. Ethan Mollick, in Named commentary item 1, is the only outside view, and Mollick had early access. The facts in main-line item 3 rest on only two newsrooms, the New York Times and Bloomberg; Investing.com relays the Times and counts as the same outlet. Main-line item 4 and Product moves item 2 come from OpenAI's own blog.
Tracked sources. The long-term roster currently has 529 named sources: 302 on social platforms, 90 shows, 51 news outlets, 48 blogs, 48 paper authors and 46 newsletters, with the rest spread across earnings reports, keynotes and other channels. The same source may be tracked on several channels, so the channel figures add up to more than 529. These are counts of tracked sources, so they can't be added to or compared with last night's piece counts above. Representative names include Daniel Kokotajlo, Sergey Levine, Arvind Narayanan and Chris Lattner on social platforms; Terence Tao and Lilian Weng among blogs; and Sam Altman, Dario Amodei and Demis Hassabis among show interviewees. This issue uses 26 outside sources in the body. Only outside links actually cited in the body are counted; these 26 are a separate count from the 529 tracked sources. Sources from Taiwan total 195, with 43 added in October; Taiwanese media, Taiwanese companies' official channels and local reporters in Taiwan count, while foreign outlets relaying Taiwanese media do not.