AI Morning Brief · Nov 7, 2026

Coverage window: 2026-11-06 07:00 → 2026-11-07 07:00 (Beijing Time)

Research note: English edition written primarily from English-language sources; some items cite Chinese-language sources.

Top Stories

1.OpenAI formally launches GPT-6 Cyber, raising its cybersecurity rating to "Critical"

the model behind the 948 Linux kernel flaws found on Nov 5; Reuters reported on Nov 5 that OpenAI raised its cybersecurity classification for GPT-6 Cyber to "Critical" under its safety framework, up from "High" for GPT-5.6 Cyber; OpenAI says it scores 48.5% on ExploitGym (GPT-5.6 Cyber: 32.1%) and was used to find 81 previously unknown Firefox vulnerabilities Mozilla classified as high severity; it is available first to approved Daybreak Red customers via API and Codex, expanding to all approved customers over the coming weeks, while GPT-5.6 Cyber is now available to all Daybreak Red customers (Nov 5)

2.Thinking Machines Lab nears a $5 billion round at a $40 billion valuation and releases Inkling 2 Small

Business Insider reported on Nov 6 (Dealroom, SiliconANGLE and Techstrong corroborating) that Mira Murati's Thinking Machines Lab is close to a $5 billion round led by Accel at a $40 billion pre-money valuation (about $45 billion post-money), more than tripling the roughly $12 billion valuation of its $2 billion seed in July 2025, with a close possible within days; the company declined to comment and the round is not closed; a September report that NVIDIA might fund about half remains unconfirmed; the same day the lab released Inkling 2 Small (Nov 5): a 594B-parameter mixture-of-experts model activating 17B per token, scoring 31 on the Artificial Analysis Intelligence Index, available now on its Tinker platform ($0.30 per million input tokens, $1.00 output), with full weights promised under Apache 2.0 next week after external safety testing — the weights are not out yet (Nov 5–6)

3.Alibaba teases the Qwen4 series: four tiers, flagship possibly over 3 trillion parameters (speculation)

on the evening of Nov 6 the official Qwen account posted a teaser image quoting Zhuangzi's "In the northern darkness there is a fish" (the blank, "Kun", puns on Qwen), confirming the Qwen4 series is coming; multiple Chinese outlets report four tiers — Max (flagship), Plus, Flash and a locally deployable 27B — rolling out through November and December; a flagship above 3 trillion parameters is industry speculation (Qwen3.8 Max is 2.4 trillion, Kimi K3.1 is 2.8 trillion) and Alibaba has published no benchmarks, prices or firm dates; Qwen3.8 Flash Next, already public, carries the Qwen4 architecture (new attention mechanism, N-gram memory tables, Muon optimizer) (Nov 6)

4.Anthropic's IPO timetable slips as investors balk at the $1.5 trillion valuation

Reuters reported on Nov 5 and the Wall Street Journal (via Invezz, Nov 6) that prospective investors question whether Anthropic can start its roadshow in the week of Nov 9 and list in November, calling the discussed $1.5 trillion valuation too high — one investor put fair value near $1 trillion — and seeking a bigger discount; the deal may be delayed, repriced or shrunk from a planned roughly $100 billion raise, against a tech selloff, geopolitical uncertainty, quarterly losses and large AI infrastructure commitments; Anthropic declined to comment; the S-1 was filed publicly on Oct 26 and amended on Nov 2, with the pricing range and share count still blank; the Journal put its revenue run rate above $80 billion at the end of October; Polymarket gives a 29% chance of no IPO before Nov 30 (market-odds figure) (Nov 5–6)

5.Chip stocks rebound sharply on Nov 6, partly recovering Thursday's rout

at the Nov 6 close (Investopedia, Morningstar), the Philadelphia Semiconductor Index (SOXX) rose 3.5% after Thursday's 12% plunge, NVIDIA climbed 2.8% to $186.64, AMD soared 11%, Intel added 8% and Micron rose 5%; the Nasdaq gained 1.1% and the S&P 500 0.3% while the Dow shed 0.5%, and the major indexes still posted weekly losses; the rebound is attributed to bargain hunting, not a policy breakthrough; Asian chip stocks fell the same day — TSMC, Samsung and SK Hynix all down, with the Nikkei and Kospi each off 6.0% (TSMC's reported decline varies, 4.5%–10%) (Nov 6)

6.Google releases Gemini Omni 1.2 Flash: 15-second video, 30% cheaper

Google's official blog on Nov 5: Gemini Omni 1.2 Flash is generally available across the Gemini API and Gemini for business, with native video generation and editing up to 15 seconds (from 10), durations selectable in one-second steps from 3 to 15 seconds, multi-turn conversational editing and stronger consistency; 720p pricing falls to $0.07 per second, a 30% cut from Omni 1.1; requests to the 1.1 model ID will auto-route to 1.2 from Dec 3 (Nov 5)

7.Anthropic launches Claude Personal Finance in beta: read-only bank access

Anthropic announced on Nov 5 that US Pro, Max and Ultra subscribers can link bank, credit card, loan and investment accounts via Plaid and ask Claude questions about their own money, with balances, credit limits and up to two years of transactions readable; access is read-only — Claude cannot move, manage or invest money — and chats in the hub are not used for training; the beta rolls out in waves this week, US only, up to 16 banks, and excludes Free, Team and Enterprise plans (Nov 5)

8.NVIDIA open-sources Cosmos 3 world models: four 16B checkpoints on Hugging Face

the Hugging Face blog on Nov 5: NVIDIA released four post-trained checkpoints of its Cosmos 3 family, each a post-trained Cosmos 3 Nano with 16B trainable parameters under the OpenMDW1.1 commercial licence — Transfer-Auto (multi-view driving data plus LiDAR generation), Sim-Auto (streaming synchronised camera and LiDAR), Sim-Transfer (simulator controls to photorealistic video) and Sim-Bimanual (video prediction from bimanual robot actions) (Nov 5)

9.Democrats plan a wave of AI bills when they take Congress in January 2027

Reuters reported on Nov 6 (via Devdiscourse and Crypto Briefing) that Democrats plan legislation in three directions: protecting consumers from data-center-driven electricity price rises, requiring independent safety testing of frontier models, and clarifying company liability when AI products cause harm; the leads are Senate majority-leader candidate Schumer (electricity first), Rep. Trahan (independent evaluation plus liability) and Rep. Lofgren; the bills are still being drafted, with no text or vote counts, and a Trump veto remains the main obstacle; separately, Senator Sanders on Nov 4 reposted his call for a national data-center moratorium, and 40 of 41 local referendums passed limits, though a federal moratorium remains far from law (Nov 4–6)

10.Moonshot AI open-sources Kimi K3.1 Pro weights: 2.8T parameters, 1.56 TB

as promised on Oct 23, Moonshot released the full weights on Hugging Face on Nov 4 UTC (Nov 5 Beijing time): 2.8T total parameters with 104B active per token, 1M context, 96 safetensors files totalling 1.56 TB, MXFP4 weights, under a custom MIT-like licence that requires a separate agreement for model-service providers with over $20 million in annual revenue; the company cites a Terminal-Bench 4.0 score of 65.7%, close to Claude Opus 5.5's 66.4%; it is the first open-weight release above the 2-trillion-parameter threshold since the interim export-control registration took effect on Oct 16, and whether Moonshot completed the registration is undisclosed — an unresolved grey area in analysts' reading (Nov 4–5)

Platform & Model Tracking

OpenAI (GPT/ChatGPT)

GPT-6 Cyber formally launched, with its cybersecurity rating raised to "Critical" (see Top Story 1). OpenAI announced on Nov 4 it will acquire Pierre Computer Company (a 10-person startup): its open-source Trees/Diffs libraries are already used in Codex for file browsing and code review, and its paid Code Storage will be folded into Codex, with the open-source projects staying open; price undisclosed, pending regulatory approval. Single source: Images 2.6 launched on Nov 4 (GPT-Image-2.6 Flare and Sunburst), about 50% faster, with GIF animation, image annotation and SVG export controls, live in ChatGPT, Codex and the Image API; OpenAI staff said on Nov 6 that ChatGPT and the API now generate over 4 billion images a week, up from 3 billion eight weeks earlier; help documentation also says ChatGPT and Codex will share one token quota from December (Plus/Pro/Business) — reported only by Medium and geoffreychen, not cross-checked. CFO Sarah Friar told CNBC on Nov 6 that OpenAI will press ahead with US data-center expansion despite the midterms and is raising new funds, with Q3 business up 70% quarter on quarter and enterprise demand doubling (Reuters). No new GPT chat-model version announced in the window.

Anthropic (Claude)

Claude Personal Finance launched in beta (see Top Story 7). Capital: the IPO timetable is slipping as investors question the $1.5 trillion valuation (see Top Story 4). METR's independent assessment (Nov 5), based on 11 weeks inside Anthropic (Aug 12–Oct 29), estimates model capability advanced about 1.9x faster than before October 2025 (95% confidence interval 1.5–2.8x), with about 80% of the acceleration subjectively attributed to AI itself; Anthropic redacted most key evidence and METR had no access to unreleased models, so the conclusion carries large uncertainty. The S-1 pricing range and share count are still blank, and there is still no public reply to Senator Warren. Single source: Claude voice mode is reportedly in development for launch as early as this month (Gadgets360), and Message Threads are in public testing on the Claude platform, cutting latency on long agent sessions by avoiding resending full conversation history.

Google DeepMind (Gemini)

Gemini Omni 1.2 Flash released with a 30% price cut (see Top Story 6). No new Gemini model, pricing or availability announcement in the window; the public launch date for Gemini 4 Carbon for most customers is still unconfirmed; no new development on the K-12 schools policy.

Meta

On Nov 5 Meta pulled the preview links for its proposed commercial terms and indefinitely paused four updates, including the Meta Commercial Terms that were due to take effect on Nov 20; existing terms stay in force; courts in Norway and Brazil had issued preliminary injunctions and the UK CMA has opened an investigation into possible abuse of market dominance; the terms had been criticised as allowing scraping of Meta Business account users' website content (NiemanLab). The Conductor cloud timetable is unchanged: read-only on Nov 13, shutdown and data deletion from Nov 20. No new Llama or Muse model, pricing or availability release in the window; the reported Fourier humanoid robot prototype remains single-source.

xAI (Grok)

Under the EU AI Act's Article 50, from Nov 7 the API output of grok-4.7 / 4.6 / 4.5 / 4.3 and three grok-4.20 variants carries an invisible text watermark by default (covering text, structured output and tool-call parameters, with per-model opt-out for team admins); Anthropic is likewise extending Claude text watermarks to older models including Opus 4.7 / 4.6 and Sonnet 4.6, with Bedrock rollout due before Nov 10; the grace period for already-deployed generative systems ends Dec 2 — reported via Medium, aiearlyedition and llmmatchmaker citing xAI's official documentation, single-source reporting. No Grok 5 release or pricing change in the window.

Microsoft (Copilot/Azure AI)

Bloomberg reported on Nov 5 (via Reuters) that Microsoft is tightening oversight of neocloud providers Nebius, IREN and Nscale, exiting projects seen as too risky and debating whether to sign new neocloud deals, favouring self-built data centers instead — including not exercising a Texas expansion option with Nscale and exiting an anchor project in Loughton, UK, worth about $3 billion; Nebius and IREN fell over 13% that day and CoreWeave about 11%. Bloomberg reported on Nov 6 that Microsoft has been winding down rented Oracle compute over recent months (a 2023 deal that powered AI Bing), as the older-generation NVIDIA chips no longer meet its needs; Microsoft declined to comment and Oracle said expiring Q1 capacity was renewed or resold at 20% above the original price. Single source: Copilot Business is now included in all Partner Center packages at no extra charge, from $350 a year (aibriefing.dev). No new Copilot or Azure AI model, pricing or availability announcement in the window.

NVIDIA

Cosmos 3's four 16B open weights released (see Top Story 8). Single source: Vera CPU benchmarks (88-core Olympus, on sale Mar 16, 2026) — in a 100 MW data-center scenario, 1.6x more agents per watt than the fastest x86 and 2x agent speed versus the most efficient x86, with Perplexity reporting 1.5x faster coding (unite.ai). Reuters reported on Nov 5 that the Republican chairs of the House Foreign Affairs Committee, the Select Committee on Strategic Competition with China and the Intelligence Committee wrote to CEO Jensen Huang requesting a closed briefing on Nov 18 about a New York Times report on his ties to Chinese businessman Alice Huang, accused of passing information to Beijing, and NVIDIA's counter-espionage procedures; NVIDIA did not comment on the letter, saying it shares only ordinary business information with Asian cloud customers. No new chip release in the window; the claim that about 600,000 H200 chips are awaiting Chinese import approval in Hong Kong appears only in a UOB research note citing Reuters (see Unverified Items).

Mistral

On Nov 5 Mistral expanded Mistral Studio — managed Agents and a Python Agents SDK entered public preview, with the same code runnable locally, self-hosted or on Mistral Cloud, plus persistent sessions and resumable runs; a visual Workflow builder, Apps CLI, evaluation tools and enterprise connectors were added, built on Mistral Large 4 ("Le Chonk", 1T parameters); private-cloud and on-premises deployment is due at the end of November; named production customers include BNP Paribas, ABANCA and CMA CGM (aiweekly and dev.to reporting). No new model release or pricing change in the window; no new word on the Large 4 benchmark dispute.

China AI Companies

Alibaba (Qwen)

The Qwen4 series was formally teased (see Top Story 3). Single source: Apple Intelligence in China will run on Qwen — Apple and Alibaba announced on Oct 31 that it will reach Chinese users in December with iOS 27.3, with Siri, writing tools and visual intelligence wired to Qwen (able to answer questions about photos and PDF documents), open in the iOS developer beta from Nov 4; the feature is optional, Alibaba says user requests and generations are not used for training, and it does not include the next-generation Siri AI (zeniteq and deyronlabs reporting, with no official announcement text checked). Qwen4 still has no release, benchmark, price or firm date.

Moonshot AI (Kimi)

Kimi K3.1 Pro weights open-sourced (see Top Story 10). Single source: the November SuperCLUE leaderboard (updated Nov 5) puts Kimi K3.1 Pro first among Chinese models at 61.46, with Zhipu GLM-5.3 third (50.66) and GLM-5.3-Flash fourth (47.18), Qwen3.8-Max and Tencent Hy4-Preview tied fifth, and Doubao Seed-2.1-Pro, MiniMax M3.1/M3 and StepFun Step-5-Preview all in the top ten; the top two overall are Gemini-4-Argon (73.62) and GPT-6.1-Sol (71.76), listed for reference only (llms.blog). No new development on the distillation allegations in the window.

Tencent (Hunyuan/Yuanbao)

Follow-up: Hy4 was formally released on Nov 4 (Tencent's official blog): 770B total parameters with 49B active, 1M context, the first Hunyuan flagship with visual understanding (images, charts, visual maths); API pricing is $0.834 per million input tokens and $2.501 output, half price until Dec 31; weights and the system card are promised "in the coming days"; Tencent says it leads open-source models in blind financial-task evaluation and that training data was filtered for offensive cyber and biological content. This adds official detail to the pricing and rollout reported in the previous issue. Hy4 weights, system card and full-release licence are still unpublished, with no date (see Unverified Items).

Baidu (Ernie)

Baidu open-sourced ERNIE 5.5 Flash Base on Nov 5: a sparse mixture-of-experts with 280B total parameters and 14B active per token, pre-trained from scratch on 33.5T tokens, with 128K training context and up to 1M for fine-tuning and inference; a pure pre-trained base checkpoint with no mid- or post-training, in FP8 and BF16 on Hugging Face and ModelScope; the licence is a custom agreement that excludes the EU, UK and South Korea, and commercial use above $50 million in annual revenue needs written authorisation (The Decoder, OSChina). The full ERNIE 5.5 Flash follows later.

DeepSeek

Single source: operating figures for Harness, its agent tooling (raisink and superbuddy) — 1.54 million daily active users as of Nov 4, about 92 trillion tokens consumed daily and about 35,000 third-party plugins; the "creation mode" launched on Nov 2 had about 3.96 million views on its first Bilibili demo video by the morning of Nov 6, and the company is preparing a plugin marketplace with token incentives for creators. WIRED reported on a Vals CyberOffense evaluation (Nov 6): simulating enterprise network attacks, DeepSeek V4.1 Flash 1017 completed 73% of attack objectives on average, close to OpenAI's restricted Daybreak model (81%), and beat Daybreak in the Nova scenario; per-attack token cost was $2–3 for DeepSeek versus $17–46 for GPT-6 Sol; Kimi K3.1 Pro was also tested (letsdatascience and dealroom reporting on WIRED). No new model or product release in the window; the reported ¥105.5 billion second funding round is unverified (see Unverified Items).

Zhipu (GLM) / MiniMax

Single source: Hong Kong-listed Chinese AI model stocks fell sharply in early trading on Nov 6 — Zhipu down 15.4% to a nine-month low and MiniMax down 12.2% — reportedly tied to the AI selloff and doubts over Anthropic's IPO roadshow (TradingView citing a Reuters flash). No new model or pricing release in the window.

ByteDance (Doubao)

Single source: TRAE accounts will auto-upgrade to Doubao–Feishu accounts from Nov 18 (announced Nov 6) — code, projects, history and membership benefits carry over, and one account will log in to the Doubao app, Feishu, PICO and Volcengine; TraeWork and TraeCode were merged into the new TRAE on Oct 9 (noustime). No new in-house model release in the window.

Sand.ai

Sand.ai announced on Nov 5 a $200 million Series A+ round, co-led by CStone Capital and Wangsu Technology — its third round of 2026, about $300 million raised in ten months, at a valuation above $1 billion; its MAGI-2 Preview is a roughly 114B-parameter mixture-of-experts video model, with the full MAGI-2 coming soon, and its VidMuse product passed $10 million in annual recurring revenue two months after launch (TechNode Global).

Other China developments

The Wuzhen Summit closed on Nov 5 with the first Global Open AI Action Initiative — 18 actions across open development, safety and trust, dialogue and governance, and inclusive benefit — plus a self-regulation convention for AI agents proposing layered, verifiable agent identity and permission management. Security (single source): Check Point disclosed on Nov 6 that the AI penetration-testing platform ARTEX was abused by criminals to automate intrusions, stealing over 190,000 game-account credentials and more than 300 Apple code-signing certificates; one operator used DeepSeek V4.1 Flash and Kimi K3.1 Pro to process about 36 billion tokens, with about 3.3% of credentials (around 6,000) still valid, at a five-day running cost of only about $180–360 (cybersecuritynews).

Policy and regulation

China's NDRC full-lifecycle AI law (announced Nov 3) saw no new official development in the window. China's draft AI export-control rules (comments open until Nov 15) saw no new official action; Nov 4–5 analysis argues APIs, SaaS and already-public open weights fall outside the controls and that Kimi K3.1 Pro's on-schedule release shows frontier models can still ship under the interim registration regime, though whether Moonshot completed registration is undisclosed — an unresolved grey area. The Democrats' AI bill plans (see Top Story 9). The US Department of Justice formed a Super Intelligence Task Force (DOJ-SITF) on Nov 5, co-chaired by Deputy Attorney General Stanley Woodward and Principal Deputy Associate Attorney General Stephen Weiss, coordinating criminal and civil enforcement against foreign actors, transnational criminal organisations and AI fraud, absorbing the AI litigation task force set up in January; it creates no new offences, does not target AI companies that break no law, and will liaise with the White House super-intelligence task force (chaired by Jay Clayton); Nov 28 is the executive-order deadline for legislative proposals on a federal definition (PYMNTS, Nextgov).

Capital markets

Thinking Machines Lab's reported $5 billion round (see Top Story 2). Anthropic's IPO under pressure (see Top Story 4). Chip stocks rebounded on Nov 6 while Asian chip stocks fell (see Top Story 5). Sand.ai's $200 million Series A+ (see above). Zhipu and MiniMax fell sharply in Hong Kong (see above). Conflicting Perplexity funding figures continue (see Unverified Items).

Worth Reading

1."How the Midterms Raised the Stakes for AI Regulation" — Best Law Firms, Nov 6, 2026

how the midterms changed AI regulation — 109 state AI laws passed in 29 states in the first half of 2026 plus 28 data-center laws, 40 of 41 local data-center referendums passing limits, and federal pre-emption stalled — the context for this issue's Democratic bill plans and the data-center moratorium debate

2."Daily Chip, 5 Nov: The market stopped buying the whole AI story" — Silicon Brief, Nov 5, 2026

a line-by-line breakdown of the Nov 5 rout — who still grew (Coherent's data-center business, +79%) and who got squeezed (memory price rises hitting phones) — arguing the bottleneck has moved from chips to permits and power

3."Your AI Text Is Now Watermarked: What EU AI Act Article 50 Means" — Medium, Nov 6, 2026

against Article 50 and the Dec 2 transition deadline, how Claude's and Grok's watermarks differ in visibility, opt-out and detection rights — the most practical read for anyone shipping AI products

Unverified Items

1.

DeepSeek's reported second funding round: Trivium China tracking (via a caybles roundup) puts it at about ¥105.5 billion ($15.8 billion) from nearly 30 investors, about 27% state capital, at a valuation of about ¥615.5 billion ($92 billion), which would be the largest single round in Chinese AI history — only that roundup reports it, with no company announcement or mainstream confirmation, so it is not adopted.

2.

Conflicting Perplexity funding figures (carried over): DefiLlama still records a Nov 2 Series F-1 of $3 billion at a $30 billion valuation (updated Nov 6), while several Oct 31 checks say the last confirmed valuation was $20 billion in September 2025 and anything above $30 billion was only under discussion in August, with no company announcement or mainstream confirmation — the $3 billion / $30 billion figures are not adopted.

3.

About 600,000 NVIDIA H200 chips reportedly awaiting Chinese import approval in Hong Kong: only a UOB research note citing Reuters (two sources), with no Reuters original found — unverified.

4.

Zhipu and MiniMax share falls in Hong Kong: only a TradingView citation of a Reuters flash, single source, not cross-checked.

5.

Apple Intelligence in China entering developer testing: only zeniteq and deyronlabs aggregators, with no Apple or Alibaba announcement text checked against.

6.

OpenAI Images 2.6 and the shared ChatGPT/Codex token quota from December: only Medium and geoffreychen reporting, with the help-document wording not cross-checked.

7.

Anthropic's Claude voice mode and Message Threads public testing: only a Gadgets360 report, with no official confirmation.

8.

Microsoft folding Copilot Business into Partner Center packages: only aibriefing.dev, single source.

9.

Broadcom seeking packaging capacity beyond TSMC: only a supply-chain report (eastasiabrief) saying TSMC supplies about 95% of front-end wafers and CoWoS lead times run 52–78 weeks, not cross-checked.

10.

Micron's Taoyuan plant reportedly halted: Bloomberg says Micron has largely paused construction of its $3–4.5 billion Taoyuan plant (a labour dispute), with Reuters noting it could not independently verify and Micron saying existing operations are normal.

11.

Oracle's New Mexico Project Jupiter (2.45 GW) facing a worse outlook after Haaland's election: a ZeroHedge Nov 6 article summarising older FT/Bloomberg reporting, second-hand and not cross-checked.

12.

The reported "tenant must use only Google-approved TPUs throughout" clause in Google's TeraWulf backstop appears only in an 8-K exhibit reading (Mosaic Singularity); CoinDesk and Reuters did not mention it — authenticity awaits checking against the 8-K itself (carried over).

13.

MIT weight-space programming: the blog post is not a peer-reviewed paper, the headline math gain came mainly from fixing a looping defect rather than a general capability gain, and Willison's reproduction was a single self-reported experiment — conclusions need qualification (carried over).

14.

Samsung's second Texas plant and AMD/NVIDIA orders: the only source is South Korea's Maeil Business Newspaper via Reuters, with none of Samsung, AMD or NVIDIA confirming; the 2030 production timetable is reported only (carried over).

15.

Salesforce's reported $2.7 billion price for Cresta: Salesforce has not disclosed it; only The Information reported it and other outlets flag it as unconfirmed (carried over).

16.

Grok Bot phone calls, the Claude Code cost optimisation, OpenAI Daybreak Red in GovCloud and Cat Wu's move were each reported only by The Neuron daily or a single source and have not been independently cross-checked — treat as secondary items (carried over).

17.

NVIDIA's reported "49% cap on China customers": still unconfirmed in follow-ups; not a fact in the main text (carried over).

18.

Anthropic's reply to Senator Warren: still no public reply after the Nov 3 deadline; remains pending (carried over).

19.

Tencent Hy4 weights, system card and full-release licence: Tencent promises only "in the coming days", with no date; the full-release licence terms are unpublished and the preview's Apache 2.0 does not carry over (carried over).

20.

OpenAI's reported up-to-$30 billion funding (about $1.25 trillion valuation, UAE investors leading): media-reported only; the official Abu Dhabi four-agreement announcement disclosed no amounts or financing terms (carried over).

21.

Anthropic's revised S-1 Q3 revenue of $16.96 billion and $895.4 billion in compute commitments: only an AI Briefing Nov 3 summary; the SEC filing has not been checked item by item (carried over).

22.

GPT-6.2 Astra and Claude 5.5 leaks: only a KuCoin Nov 2 single-source report (citing AIBase), with no company announcement or verifiable rollout evidence in the window (carried over).

23.

Google Gemini "hacked three companies in a security test, then stopped on its own": still only an Al Jazeera mirror site (edgeone domain), with no Wall Street Journal original or Google statement quoted directly (carried over).

24.

GPT-6 Astra Law pricing page: only Observatory Nov 1 and TestingCatalog reports, with no check against OpenAI's pricing page (carried over).

25.

NVIDIA Vera Rubin bandwidth target cut: a Ranzware Nov 1 report comparing product pages, a silent September revision rather than a Nov 5 announcement, with no NVIDIA statement to check against (carried over).

26.

Meta and Fourier humanoid robot prototype: only an Engadget report via a The Diary of AI Nov 1 sidebar, with no Meta announcement to check against (carried over).

27.

Microsoft AI PC roadmap leak: only a TechSpot Nov 4 report citing leaked slides (HXL on X), unconfirmed by Microsoft (carried over).

28.

Open questions carried over (no new development in the window): Grok 5, Anthropic's S-1 pricing range and share count (still blank), Zhipu's rumoured "3.5 trillion parameters", ByteDance / Alibaba / Tencent RTX PRO 5500 orders, Ant's Ling / Ming single-source reports, Alibaba / Tencent / ByteDance ten-year memory contracts (single source), the Gemini 4 Carbon full-availability date, NVIDIA Nemotron 3.5 Transcribe (no official blog citation), distillation allegations (DeepSeek / Moonshot vs GPT-6 Astra), Meta's parental-reminder expansion (single source), xAI's Mississippi house purchase in exchange for dropping a lawsuit (single source), NVIDIA reportedly guaranteeing Nscale (single-source analysis), Grok Imagine Video 1.5 Lite (single source), OpenAI's actual first-round fundraising total, and the Thinking Machines Lab $5 billion round (reported, not closed); Qwen4 is only a teaser in this issue — release, benchmarks, price and firm dates are still open; whether Moonshot completed export registration for the Kimi weights is undisclosed; the 948-flaw figure from the previous issue is partly confirmed by the formal GPT-6 Cyber launch (model and rating now official), but no per-CVE list has been published, so it is still attributed to OpenAI.