AI Morning Brief · Nov 14, 2026

Coverage window: 2026-11-13 07:00 → 2026-11-14 07:00 (Beijing Time)

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

Top Stories

1.OpenAI launches ChatGPT Wallet, letting its dots agents pay for purchases

OpenAI launched ChatGPT Wallet on Nov 13: its dots agents can complete shopping payments with one-time tokens, never touching the real card number, with each transaction requiring the user's own biometric confirmation (Face ID or fingerprint); launched with American Express, Mastercard and Visa, initially limited to US ChatGPT Pro subscribers aged 18 and over; OpenAI takes no cut at launch and published a Commerce Partner Playbook asking merchants not to block agent checkouts. (Reuters, Nov 13)

2.Anthropic IPO follow-up: orders reach six times the shares on offer on day three of the roadshow; pricing Nov 18, trading Nov 19

Reuters reported on Nov 13 that Anthropic's IPO orders had reached six times the shares on offer by the third day of the roadshow (about two times on day one); the price range is $70–$78, with 825M Class A shares, raising about $64.4B in the public offering (plus about $26B in a private placement), for a basic valuation of about $0.9–1.0T and about $1.4T fully diluted; pricing is expected Nov 18 and Nasdaq trading on Nov 19 under ticker ANT; Polymarket bets put the most likely first-day market value at $1.2–1.5T; the FT says bankers plan to raise the IPO price on the strong demand, targeting a valuation back above $1.5T; Barron's adds that October revenue reached an $81B annualized run rate, Q3 revenue was $17B with a 15% adjusted margin and a 35% long-term target, and up to 20% of shares are aimed at retail investors. The six-times figure comes from sources cited by Reuters; Anthropic declined to comment and order numbers may still change. (Reuters via Unite.AI, Nov 13)

3.NVIDIA CEO search and congressional summons follow-up: FT says the board approached Arm CEO Rene Haas, the board denies formal contact; Senate Banking Committee summons Huang for Dec 3 and Lutnick for Dec 10; Reuters says the White House is discussing removing Lutnick

The FT reported on Nov 13 that NVIDIA's board has approached Arm CEO Rene Haas about succeeding Huang, also naming internal candidate Ian Buck; the board said the same day that it has not formally contacted any candidate, that speculation about names is premature, and that it is selecting a search firm; CFO Colette Kress remains interim CEO; Wired reports an internal post urging the board to weigh employees' wishes has drawn more than 1,600 comments, many demanding Huang's reinstatement. Senate Banking Committee chair Tim Scott wants former CEO Huang to testify on Dec 3 and Commerce Secretary Lutnick on Dec 10; NBC reports Lutnick has agreed to appear (updated Nov 13 11:30 UTC). House committee chairs separately want Huang for a closed-door transcribed interview on Dec 1, NVIDIA general counsel Tim Teter on Dec 2 and Commerce deputy Jeffrey Kessler on Dec 3; Huang must confirm in writing by 20:00 ET on Nov 13 or face a subpoena; House Foreign Affairs chair Mast says Huang has agreed to testify before Thanksgiving on Nov 26, with the date still being negotiated. Reuters reported on Nov 13 that senior White House officials are discussing whether to force Lutnick out for not doing more to restrict chip sales to China; the White House and Commerce Department did not respond to requests for comment; Treasury Secretary Bessent told CNBC on Nov 13 that Treasury, State, the War Department and USTR have for months favored putting Singapore cloud vendor Megaspeed on the Entity List, but the decision rests with Commerce, and he questioned whether NVIDIA's customer due diligence was adequate. Note: White House officials said on Nov 12 that Trump has "full confidence" in Lutnick, so the two signals conflict. (FT via Reuters, single source on the Haas approach; Reuters on Lutnick)

4.NVIDIA follow-up: shares fall another 2.01% to $162.85 on Nov 13 as HSBC cuts the stock to Hold with a $170 target; chip stocks slide broadly and AMD gives back its gains

NVDA closed at $162.85 on Nov 13, down 2.01% (prior close $166.19), after an intraday low of $158.39, on volume of 360M shares; HSBC cut the rating to Hold and slashed its target price from $375 to $170; the stock had rebounded more than 2% in early trading; the next tests are Q3 earnings on Nov 17 (Kress's first earnings call) and an all-hands on Nov 18. The Philadelphia Semiconductor Index fell 3.43%; AMD closed at $534.06 (-8.75%) and Broadcom at about $346.71 (-4.84%), roughly erasing their Nov 12 gains (AMD +5.86%, Broadcom +5.54%); TSMC's ADR closed at $398.73 (-1.67%). (Finnhub data; Motley Fool, Nov 13)

5.Microsoft 365 global outage follow-up: largely resolved on the afternoon of Nov 13; Microsoft blames a network-capacity upgrade, not an attack

Microsoft said on the afternoon of Nov 13 that it had fixed the root cause and service was "broadly restored and stabilizing", with the service health page showing all services "running normally", though mail queues were still draining and some users still saw delays; Downdetector's Outlook reports fell from a peak of about 130,000 to about 1,600; Microsoft said the incident was "not related to the network or AI/agents" and stemmed from "an issue introduced during an earlier network capacity upgrade that surfaced after a recent network change", promising a full post-mortem; the outage ran from about 16:00 GMT on Nov 12 for more than 24 hours, affecting 74 countries and regions, hitting Outlook, Teams, Copilot, SharePoint and OneDrive; it is Microsoft's second service outage this year. (USA Today and MacRumors, Nov 13)

6.DeepSeek opens its full app-layer API, letting developers clone the DeepSeek app one-to-one

DeepSeek announced on Nov 13 that it is opening its full application-layer API, folding the app's web search, deep-thinking search, file input, link reading, voice interaction and safety checks into its existing API; it also opened standalone Search, Fetch, OCR, ASR and TTS tool interfaces and published an open-source frontend demo (cdn.deepseek.com/chat-demo); the new capabilities run on deepseek-flash (DeepSeek-V4.1-Flash-1017), and extended tool calls are charged on top of token fees. (DeepSeek's official X account via Crypto Briefing; ITHome, Nov 13)

7.Tencent's Hunyuan launches Hy Realtime, a real-time voice model that can sing and call tools

Tencent's Hunyuan team launched the Hy Realtime voice model on Nov 13, billed as "listen, think and speak at the same time": it understands vague, half-finished or mid-sentence-changed speech, and can search the web and call tools based on conversational intent, plus a new singing ability (built with Tencent Music's Lingyin AI team, with song requests by singer, genre, language or year); it is live in the Yuanbao app's voice calls (the "Chuyu" voice) and Tencent customer service, with a Tencent Cloud API coming soon. (ITHome and other Chinese outlets, Nov 13)

8.Zhipu and MiniMax follow-up: Hong Kong shares plunge for a second straight day, Zhipu down 10.04% and MiniMax down 8.19%

Hong Kong closed lower on Nov 13, with the Hang Seng down 1.55% and Hang Seng Tech down 3.02%; Zhipu (02513.HK) fell 10.04% (about HK$408.4 by calculation from the Nov 12 close of HK$454; no source directly stated the closing price), with turnover of HK$5.178B and a 4.67% turnover rate, the highest among newly listed stocks; MiniMax (00100.HK) closed at HK$152.4, down 8.19% (prior close HK$166), with turnover of HK$2.257B; both fell more than 8% for a second consecutive session, with Zhipu down more than 12% intraday; the market linked the AI sell-off to the news of NVIDIA CEO Jensen Huang's resignation. (21st Century Business Herald; Finnhub, Nov 13)

9.US October PPI rises 6.5% year over year, the fastest since 2022

The US Bureau of Labor Statistics reported October PPI on Nov 13: up 0.3% month over month (expected +0.4%) and up 6.5% year over year (September was +6.3%, expected +6.6%), the fastest year-over-year pace since 2022; core PPI rose 5.0% year over year and 0.3% month over month; goods rose 0.5% and services 0.2%; markets still largely price a 25bp Fed rate hike in December (CME FedWatch had been about 90%), with some discussion of a possible 50bp move. (BLS data via TTNews and Business of Fashion, Nov 13)

10.Mistral publishes the ML4 technical report: 1 trillion parameters, the best open-weight model on several benchmarks

Computer Weekly reported on Nov 13 that Mistral has published a 30-page technical report for Mistral Large 4 (ML4): 1 trillion parameters (52B active per token), 19.7% on FrontierCode 2.0 and 41% on Terminal-Bench 4.0 (up 13 points after three extra weeks of RL, with 30% fewer output tokens), the best among open-weight models on both; the three safety benchmarks (JailbreakBench, StrongREJECT, AgentHarm) average 96.6% for refusing or safely handling harmful requests; it was trained for 45 days on 3,840 NVIDIA GB200 chips in a Paris data center; CrowdStrike, Elastic and Rubrik have tested it; Cloudera also announced on Nov 13 that it is expanding its enterprise AI ecosystem to include Mistral and three other partners. (Computer Weekly, Nov 13)

Platform & Model Tracking

OpenAI (GPT/ChatGPT)

ChatGPT Wallet launched (see Top Story 1). Follow-up: the Cursor cutoff did not take effect and access was temporarily extended — on Nov 12 OpenAI appended an update to its August announcement: "We have temporarily extended Cursor's access to our models beyond November 12", with no new deadline given; as of Nov 13 Cursor still lists GPT-5.6 Sol/Terra/Luna, with no GPT-6 models; Cursor has made no public statement. GPT-6.2 remains unreleased, with only leaked model IDs and screenshots.

Anthropic (Claude)

IPO orders reached six times the shares on offer on day three (see Top Story 2). Single source: the seven co-founders are selling about $1.75B of stock combined — per the prospectus, each of the seven co-founders is selling 3,378,378 shares, 23,648,646 in total, worth about $250M each and about $1.75B in total at the $74 midpoint, about 1.5% of each person's holdings; JonesTrading is also underwriting a block trade of up to $500M (a Pollar roundup citing Bloomberg and FT, unverified; the piece was published Nov 10 and updated Nov 13, so the timing boundary is unclear). In a live conversation at JPMorgan headquarters on Nov 13, CEO Dario Amodei told JPMorgan CEO Jamie Dimon that "compute is the deepest business dilemma" the company faces, saying the company wants to buy compute cheap and top up on the spot market in peak season; the S-1 shows total compute commitments of $900.9B through 2045, about $451.6B of it due before 2030; Amodei also stressed safety and a "no jerks" culture, saying he spends 40–50% of his time on people and culture. Claude failed twice on Nov 13 — the status page records two incidents: a login authentication failure from 09:58 UTC (claude.ai, Claude Code, Console, Cowork), resolved 11:58 UTC, and a performance degradation from 13:09 UTC (claude.ai, Claude API, Claude Code, with connectors, meetings and Managed Agents affected), marked resolved at 15:38 UTC; the company gave no cause; some reports linked it to the Microsoft 365 outage, but Anthropic mentioned Microsoft only in connection with the Nov 12 Microsoft 365 connector incident. Self-hosted Claude Managed Agents entered private beta: announced Nov 12, enterprises can run the agent framework, sandboxes, credentials and session history in their own cloud or on-premises, with model calls still going to Anthropic by default or to Bedrock, Google Cloud or Microsoft Foundry; no public pricing or GA date, available only through customer teams. No new developments in the South Korean PIPC investigation; Sonnet 4.5 deprecation unchanged (deprecated Sep 30, hard retirement Nov 30, with Sonnet 5.5 as the recommended replacement).

Google DeepMind (Gemini)

Gemini 4 Hydrogen rollout details (follow-up) — Google's official blog (published Nov 12, updated Nov 13) confirms Hydrogen is now live globally: across AI Studio, the Gemini API, the enterprise edition, the Gemini app, Google Search and Antigravity for Pro/Ultra subscribers; it replaces Gemini 3.8 Flash in Search; four reasoning levels (low/medium/high/xhigh); at the default high setting it scores 51 on the Artificial Analysis Intelligence Index (above Argon at 50) and 54 at xhigh; Terminal-Bench 4.1 is 57% (3.8 Flash was 18.2%), with tasks 2.5x faster; pricing is $1.50/$7.50 per million tokens, flat across the full 1M-token context; Antigravity adds an Orchestrator that switches automatically between Carbon and Hydrogen within a session. Google AI Plus lost Gemini Pro on Nov 13 — per Google's October email to subscribers and a Nov 13 re-check, the $4.99 Google AI Plus plan can no longer select the Pro model in the Gemini app and Search AI Mode from Nov 13, with the plans page now headlining Gemini 4 Hydrogen; keeping Pro requires upgrading to the $19.99 AI Pro plan; existing Pro conversations are not deleted and continue under other models. AI Plus will also lose Claude in Antigravity on Nov 18.

Meta

Muse can now make outbound calls for all US paid subscribers — Meta's chief AI officer Alexandr Wang announced the outbound-calling feature has expanded from a small September test to all US paid subscribers, letting Muse call restaurants, insurers, cable customer service and the DMV on the user's behalf; the base paid Muse plan starts at $20/month (announced Nov 12 US time, about Nov 13 Beijing time, on the window's edge). Follow-up on the five-company lawsuit seeking a freeze on frontier models (no new ruling): media on Nov 13 added details of the Nov 12 suit — the plaintiffs are the Irreplaceable Education Fund, Public Citizen, the Heat Initiative and the Alliance for Secure AI, represented by former New Jersey attorney general Matthew Platkin, suing Anthropic, Google, Meta, OpenAI and SpaceX Super Intelligence in San Francisco Superior Court to bar development and deployment of any model beyond the strongest systems at the time of filing, and to bar AI from autonomously researching its successors, under California's unfair competition law and public nuisance; the defendants have not responded and there is no freeze order. No new Llama or Muse model, pricing or availability release in the window.

xAI (Grok)

Grok Bot can join video meetings and pick up work from them — xAI announced on X on Nov 13 that Grok Bot can join video calls, answer questions raised in the discussion, take on assigned tasks and track meeting decisions; the feature is for Super, Ultra and all Team/Enterprise plans (single source, based on xAI's X post, unverified). Single source: The Information reports SpaceX AI cut about 60 people from the Grok Imagine team and is shifting to coding tools and agents — The Information reported on Nov 13 that SpaceX's AI division cut about 60 members of the Grok Imagine (image/video generation) team over the past week, keeping only a few to maintain the product, with the strategy shifting to AI coding tools and agents, against a backdrop of Chinese companies leading in video generation; of xAI's 25 news posts since Aug 7, none is about Imagine and 12 are about Grok Bot, supporting the shift; xAI has not responded publicly (paywalled, based only on the public summary and X posts, unverified). Security: the ArcticRAT Android trojan calls the Grok API to filter stolen SMS verification codes — security firm Zimperium disclosed on Nov 12 that ArcticRAT (78 samples found) calls Grok through the xAI API to sift one-time codes and financial information out of stolen text messages; the malware is still under active development. Single source: a fourth woman joined the CSAM class action against xAI/Grok — a Wyoming woman (Jane Doe 4 in the complaint) joined three Tennessee teenagers suing xAI, alleging her stepfather used photos of her as an 11-year-old to generate more than 7,000 sexualized images via Grok; the stepfather was arrested and later killed himself; the plaintiffs' lawyers invoke Masha's Law (single source, citing the Daily Mail, unverified). Grok 4.8 still has no official confirmation (see Unverified Items).

Microsoft

Microsoft 365 global outage resolved (see Top Story 5). No new Copilot or Azure AI model, pricing or availability announcement in the window.

NVIDIA

CEO search and congressional summons (see Top Story 3). Shares fell another 2.01% to $162.85 on Nov 13, with HSBC cutting the rating (see Top Story 4). Follow-up on the 8-K details: the SEC 8-K filed Nov 12 shows the board accepted Huang's resignation as president, CEO and director on Nov 11, effective immediately; CFO Colette Kress serves as interim CEO; Harvey C. Jones (former Synopsys CEO) was named the company's first-ever board chair; the board will search internally and externally for a permanent CEO with no set deadline; the announcement gave no reason for the departure and no investigation findings; the Hugging Face acquisition was not yet completed as of the filing. Q3 earnings are due Nov 17.

Mistral

ML4 technical report published (see Top Story 10). Mistral CEO Arthur Mensch urged "immediate action" at the EU leaders' summit: at the Nov 13 Dublin summit (the first EU leaders' summit themed on AI), Mensch and Nobel laureate Geoffrey Hinton were invited to speak; Mensch urged the EU to "act now" to avoid falling behind in the AI race, a stance contrasting with US CEOs (Altman, Amodei) who called for slowing down AI this year.

Perplexity

Teammates is now on the Mac app (Max subscribers only): Perplexity announced on Nov 13 that its proactive agent feature Teammates has expanded from iOS and the web to the Mac app, for Max subscribers only; Teammates can independently or collaboratively take on ongoing work such as QA, customer service, research and growth marketing. Cross-platform research: shopping answers from ChatGPT, Gemini and Perplexity can lead to counterfeit retailer sites, with Perplexity the worst — a study by IP-protection firm Podqi (reported by The Verge) tested 6,795 answers covering 151 brands: 15% of Perplexity's answers included high-risk sites, versus 8% for ChatGPT and 4% for Gemini; all three had cases of mistaking counterfeit sites for brands' official sites; Perplexity responded that it has been the most accurate on trusted research benchmarks for years, while conceding it is not 100% accurate. Conflicting reports on new funding continue (see Unverified Items).

China AI Companies

Alibaba (Qwen)

Qwen Code v0.27.3 (Nov 13) — a routine minor release: chips above the Web Shell composer flag unfinished background workflows, GitHub channel event prompts now support Chinese, Managed MCP limits were adjusted (32 live selections, 16 enabled), and a send-only DingTalk webhook channel was added; no model-level change. The Qwen4-Max preview is still running (Nov 11 10:00 to Nov 25 24:00), with no new details in the window; no new developments on T-Head's A1 filing.

Baidu (ERNIE)

No new releases on day two of Baidu World 2026, only recaps and analysis; the Nov 12 items (Token Exchange, ERNIE 5.5 Flash, ERNIE Lightning/Wizshot, Tianchi Supernode) were covered in the previous issue. ERNIE 5.5 Flash Preview weights remain on schedule to be open-sourced before Nov 25, with no new developments in the window.

Tencent (Hunyuan/Yuanbao)

Hy Realtime real-time voice model launched (see Top Story 7). Follow-up: Hy4 weights still not released as of Nov 13, with the official line still "planned open source" and "soon", no new developments (same status as Nov 12).

Zhipu (GLM) / MiniMax

Hong Kong shares plunge for a second straight day (see Top Story 8). The H3-Image license reading remains single source (see Unverified Items); no new corroboration for the claim of Zhipu ARR of $1.6B in the window (see Unverified Items).

ByteDance (Doubao)

Single source: AI chip chief Shi Yunfeng has reportedly submitted his resignation, with in-house chips underperforming and not deployed at scale — a Nov 11 exclusive report says ByteDance's AI chip and DPU team head Shi Yunfeng has submitted his resignation and is in handover; ByteDance's self-developed chips taped out in 2026 but underperformed internal expectations, and Ada (for Doubao) and SeedChip (for Seedance) remain in operator adaptation and have not been deployed at scale internally; Shi has been at ByteDance for eight years and has reported to Volcano Engine head Yang Zhenyuan since February 2026 (single source: 21st Century Business Herald citing ZAKER Finance and Xinghe Business Observer; ByteDance has not commented). No new developments on Zhang Yiming's 2029–2030 catch-up timeline or the Muse-style personal assistant; rumors about the departure of Seed chief Zhou Chang remain unverified.

DeepSeek

Full app-layer API opened (see Top Story 6). Follow-up on the second funding round (single-source chain): The Information reported on Nov 13 that firms linked to businessman Alice Huang — Century Litong and its Shanghai-listed parent Jiangsu Litong Electronic — took part, via an SPV set up by Monolith Management, in DeepSeek's second funding round completed in July–October (105.5B yuan raised, about 615.5B yuan / $86B post-money valuation); Litong is described as a "major black-market supplier of NVIDIA hardware", allegedly moving about 240,000 NVIDIA chips worth at least $7.5B into China since 2023; Jiangsu Litong invested about 95M yuan in the SPV. Neither DeepSeek nor Litong has publicly confirmed; the report corroborates Caixin's Nov 12 figure of about 100B yuan, with the investor-list details being new (The Information's original is paywalled; the above is second-hand relay, and the chip-smuggling allegation is unverified). A claim that DeepSeek API search is priced at "5 yuan per 1,000 calls" appears only in a single-source relay with no confirmation on DeepSeek's official pricing page (see Unverified Items). No new model release in the window.

Moonshot AI (Kimi)

No new K3/K3.1 release or developments in the window; only a Kimi API price summary page (traktoken.com, data updated Nov 13, K3 priced at $3/$15 per million tokens) and K3.1 Pro local-deployment guides, none of which is news.

Meituan (LongCat)

No new developments in the window; LongCat-2.5 weights were open-sourced on Nov 9–10 (covered earlier).

Manus

No new developments found in the window.

Other China developments

No new official action on China's AI export-control draft; the comment period still ends Nov 15 (US-side follow-ups are covered under Policy and regulation).

Policy and regulation

BAAJA follow-up — a final-vote timetable was published: the Senate is expected to hold a first cloture vote at about 18:00 on Nov 16 (on S.Amdt 6883, the BAAJA text, requiring 60 votes), a second cloture vote about 30 hours later at about 01:00 on Nov 18, then a final vote; the Nov 9 vote of 73–25 was only a procedural vote to begin debate, and the bill itself has not passed; the House must still pass it before the 119th Congress ends on Jan 3, 2027 (single source: the Hill Heat newsletter, with no confirmation against the Senate floor schedule). Neal and Wyden opened an inquiry into NVIDIA's H200 exports to China (follow-up): Representative Richard Neal (ranking Democrat on the House Ways and Means Committee) and Senator Ron Wyden (ranking Democrat on the Senate Finance Committee) wrote to Commerce Secretary Lutnick on Nov 12, opening an inquiry into the Trump administration's approval of and levy on NVIDIA chip exports to China, questioning the arrangement under which H200 chips transit the US and are exported to China after a 25% tariff as a "disguised unconstitutional export tax", and demanding preservation of all related records; they also questioned why the Commerce Department kept approving Megaspeed's purchases after the alleged diversion was known; as minority-party members they have no compulsory evidence powers; the earlier Warren/Coons letter is a separate letter. US export controls tightened further (follow-up): Senators John Curtis (Republican) and Lisa Blunt Rochester (Democrat) wrote to Lutnick asking that advanced AI chip export licenses carry conditions barring licensees from offering remote access to Chinese firms without a separate license, and limiting large exports to untrusted firms in high-smuggling-risk countries, naming Aivres (the US subsidiary of Inspur) as having exported at least $5.6B of technology to Southeast Asia in two years; separately, OpenAI has privately told members of Congress it supports the Chip Security Act, the MATCH Act and the AI OVERWATCH Act, three China-related chip-control bills (a Bloomberg Nov 12 report via WinBuzzer), to ride the annual defense authorization bill, with Senate leaders saying no final deal had been reached as of Nov 12 (the senators' letter is a single source from AI Brief; the OpenAI backing is a single-source chain). Bloomberg: banned NVIDIA chips were traced to a Shanghai data center — Bloomberg reported on Nov 13 that restricted NVIDIA chips obtained by companies founded by Alice Huang were installed in a Shanghai "Yangtze River Delta Project" AI data center: documents show 193 servers were pledged as loan collateral in January 2025 (worth about $73M), including 22 Dell servers with H800 chips (originally sold to Aperia, a Singapore NVIDIA cloud partner; Dell cut supply in February 2025 over diversion concerns) and 52 Supermicro servers with H100 or H200 chips; H100/H200 sales to China require a US license; the documents do not prove Alice Huang herself took part in the diversion; Megaspeed has bought more than $4.8B of NVIDIA chips since June 2024 (reported by Engadget and Seeking Alpha relaying Bloomberg). EU AI Act: no new action in the window; the transparency guidance C/2026/6005 (published in the Official Journal on Nov 12) and the Dec 2 new bans remain as covered in the previous issue, with no follow-up.

Data centers and energy

Nebius, CoreWeave and IREN fall again on Nov 13, with Nebius hit by several target-price cuts: Nebius closed at $115.88 (-9.92%), CoreWeave at $53.13 (-6.13%) and IREN at $18.17 (-6.72%); all three fell about 8–9.6% intraday, with the market attributing the move to doubts about AI infrastructure spending raised by NVIDIA's leadership turmoil; Baird cut Nebius's target from $340 to $200 and Truist from $355 to $245 (a MarketBeat summary, single source). The Epoch AI dataset (top data centers at 15.1 GW of compute, 38 GW by end-2028) remains single source and unverified.

Capital markets

US stocks on Nov 13 (follow-up): the S&P 500 closed at 7,561.67 (-0.40%), the Nasdaq Composite at 25,136.27 (-0.91%) and the Dow at 52,347.55 (+0.10%); for the week all three still rose 2.5%, 2.3% and 2.1% respectively; the 10-year Treasury yield closed at 5.15% (intraday high 5.17%, the highest since October) and the 30-year at 5.48%, with the VIX at about 19.3–19.79; NVIDIA closed at $162.85 (-2.01%, prior close $166.19, intraday low $158.39) and fell further after hours to about $160.93. US October PPI (see Top Story 9). Anthropic IPO (see Top Story 2). New funding this week (Crunchbase Nov 7–13 weekly list): beyond Mirendil's $1B Series A covered in the previous issue, the week's newly disclosed large rounds include Mercor's $750M Series D (at a $20B valuation, connecting AI developers with human experts for model training and evaluation), Torus's $500M growth round, Path Robotics' $300M Series E, LangChain's $300M Series C (led by CapitalG, at a $3B valuation, announced Nov 13), Nominal's $240M Series C (at a $3B valuation) and Mavria's cumulative $130M pre-seed and seed (UFORCE's $700M and Agile Robots' $500M were covered in the previous issue and are not repeated). TSMC's up-to-$1B investment in Anthropic and OpenAI: no new follow-up in the window.

Security incidents

Unit 42's new report (follow-up / new case): Palo Alto Networks Unit 42 reported (report CL-STA-1238, published Nov 12–13) that Chinese hackers used the open-source Cairn AI penetration-testing framework, connected to DeepSeek V4 Flash/Pro and Zhipu GLM 5.2, to run 25 operations against at least 19 organizations in August–September 2026, with 10 breached or having data stolen; attackers often needed only a domain name and a one-sentence objective; targets covered semiconductor manufacturing, US defense policy and civil-society organizations; one open registration endpoint leaked about 500 user records (including .mil and .gov emails); Unit 42 assesses it as one person or a very small team, with medium confidence of China-linked intelligence collection, and notes autonomous AI expands the scale and coverage of attacks, not their complexity. Separately, Unit 42's ARTEX defense guide published Nov 11 counts 1,323 hosts on the internet running CyberStrikeAI and 609 running ARTEX as of October 2026 (CybersecurityNews and Reseller NZ relaying Unit 42 — still a single-source chain). Notion/Fireworks leak follow-up: Notion's security advisory, published Nov 12 and updated Nov 13, says Fireworks had contractually promised zero data retention for Notion traffic but the investigation found it retained data from the Fireworks endpoints Notion used; the leak mainly involves Notion AI conversation records from the Insights feature, affecting under 1% of workspaces, with most records lacking author or workspace identifiers; Notion disconnected Fireworks on Nov 3, revoked the Notion API keys found in the leaked messages on Nov 11 and notified the issuers of third-party service keys on Nov 12; the Insights feature stays disabled and is no longer served via Fireworks open models; Fireworks says the attack is linked to an LLMJacking gang that profits from reselling LLM access, and that inference, the training SDK and Nexus were not found affected (second-hand summaries from ExplainX and AINotis; the original Notion advisory link was not directly obtained in this research). The Microsoft 365 outage was confirmed not to be a cyberattack (see Top Story 5). South Korea's PIPC investigation into Anthropic: no new developments in the window, and the accounts conflict (Korean media on Nov 9 said pre-review, not yet filed; a Nov 10 report said formally filed, covering 112 accounts) — see Unverified Items. Chainstack leak: no new follow-up in the window.

Worth Reading

1."Anthropic IPO could raise $64 billion as orders reach six times" — Unite.AI (relaying Reuters), 2026-11-13

The full account of the roadshow reaching six-times coverage on day three, pricing on Nov 18 and trading on Nov 19.

2."Mistral targets cyber-defence and sovereign AI with ML4" — Computer Weekly, 2026-11-13

The full reading of the ML4 technical report (1 trillion parameters, benchmark and safety scores, GB200 training details).

3."Bessent says Treasury favored limits on Megaspeed" — Reuters, 2026-11-13

The original report that Treasury favored putting Megaspeed on the Entity List months ago and that the White House is discussing removing Lutnick.

Unverified Items

1.

NVDA's Nov 13 close: Yahoo Finance and Finnhub give $162.85 (-2.01%), while some quote sources report $161.24 or $161.77, likely an after-hours or data-vendor difference; the text uses $162.85.

2.

Anthropic's total IPO raise: about $64.4B for the public offering (Reuters), with "about $90.4B" (Yahoo Finance/Bloomberg relay) and "over $100B including the greenshoe" (FT) as alternative totals including the private placement; the official filing governs.

3.

Anthropic's six-times IPO subscription: Reuters citing sources; Anthropic declined to comment and order numbers may still change; final pricing is only confirmed on Nov 18.

4.

Anthropic IPO pricing on Nov 18 and trading on Nov 19: the Reuters account; CNBC and WSJ previously reported Nov 20, the FT says pricing is expected Nov 18, and the official filing leaves the dates blank; unverified, pending the official announcement.

5.

The seven co-founders selling about $1.75B of stock: Bloomberg via Pollar, saying each of the seven sold 3,378,378 shares for about $1.75B combined at the $74 midpoint, with JonesTrading also marketing a block trade of up to $500M; single source, and the piece was published Nov 10 and updated Nov 13, so the timing boundary is unclear.

6.

Whether NVIDIA approached Rene Haas: the FT report via Reuters, which says it could not independently verify it; NVIDIA's board says it has not formally contacted any candidate.

7.

Whether Lutnick will be removed: Reuters citing sources that the White House is discussing it, with no official response; White House officials said on Nov 12 that Trump has "full confidence" in Lutnick, so the signals conflict.

8.

Huang testifying Dec 3 and Lutnick Dec 10: the Senate Banking Committee summons is a FourWeekMBA relay of NBC/CNBC, with Lutnick's agreement to appear a single-source update; the House's Dec 1 closed-door interview requires Huang's written confirmation by 20:00 ET on Nov 13 or a subpoena, and no report of the outcome appeared in the window; Mast says Huang has agreed to testify before Thanksgiving on Nov 26, with the date still being negotiated.

9.

Huang's Nov 18 closed-door briefing: no public record of a reply before the Nov 11 17:00 ET deadline, and with his Nov 12 resignation it is unclear whether the briefing will happen or who would attend; the discussion has now been superseded by the Dec 3 Senate Banking Committee testimony request.

10.

The two BAAJA cloture votes (about 18:00 on Nov 16 and about 01:00 on Nov 18): reported only by the Hill Heat newsletter, with no confirmation against the Senate's official schedule.

11.

The three senators' letters pressing Secretary Lutnick: only two secondary sources, with no confirmation from Reuters, Bloomberg or other major outlets.

12.

The Curtis/Blunt Rochester letter asking Lutnick to close the "remote access" loophole: a single source (AI Brief), unverified; OpenAI's support for the Chip Security Act, MATCH Act and AI OVERWATCH Act: a Bloomberg Nov 12 report via WinBuzzer, a single-source chain.

13.

Bloomberg's report that banned NVIDIA chips were traced to a Shanghai data center: relayed by Engadget and Seeking Alpha; the documents do not prove Alice Huang herself took part in the diversion; Megaspeed's purchases of more than $4.8B of NVIDIA chips since June 2024 are unverified.

14.

Litong moving 240,000 NVIDIA chips (worth at least $7.5B) into China: only a relay of The Information's reporting, with no other independent source, and the allegation involves smuggling.

15.

DeepSeek's second funding round: the 100B yuan (Caixin) and 105.5B yuan (21st Century Business Herald and others) figures coexist, with a pre-money valuation of about 500B yuan versus post-money 615.5B yuan; no official confirmation; The Information reported on Nov 13 that Litong-linked firms took part, still a single-source chain.

16.

Claims that companies linked to Alice Huang invested in DeepSeek and promised NVIDIA chip access: a single KuCoin aggregation flash, with no mainstream corroboration; the text itself says the intelligence link "remains only a question".

17.

DeepSeek API search pricing of "5 yuan per 1,000 calls": only an ITHome relay of eurisko.com.br, with no confirmation on DeepSeek's official pricing page.

18.

ByteDance chip chief Shi Yunfeng's resignation: a single-source report, with no response from ByteDance and no word on his next role.

19.

Zhipu's Nov 13 closing price: multiple sources give only the -10.04% decline, with no closing price stated directly; about HK$408.4 calculated from the prior close of HK$454 is a computed value, not source data.

20.

OpenAI reportedly completing a $15B round at a $1.25T pre-money valuation: an Analytics Insight weekly report on Nov 13 saying "reportedly", with no round stage or investors disclosed and no other source cross-checking; the same piece also says UK AI chip startup Fractile raised a $720M Series C (at a $7B valuation), also single source.

21.

Grok 4.8 quiet rollout: only social posts and aggregation headlines saying it reached a limited set of Bot users on Nov 12, with no confirmation from xAI.

22.

About 60 layoffs from the Grok Imagine team: The Information's article is paywalled and unread in full; only its public summary and X posts were used, and the figure of about 60 is unconfirmed by xAI.

23.

The unreleased OpenAI security model scoring 77% on ExploitGym: audience reports only, with no official post; the GPT-6 Cyber comparison figure appears as both 48.5% and 52.6% in different reports.

24.

The Unit 42 autonomous-intrusion report: a single-source chain (the report relayed by CybersecurityNews and Reseller NZ), unverified.

25.

CoreWeave's Q3 figures: single-source aggregation, with no original Reuters or Bloomberg link found.

26.

Claude Fable 5.2 fast mode on the Messages API: a single API-aggregation briefing, unverified.

27.

Mirendil's $1B Series A, UFORCE's $700M Series A, Agile Robots' $500M round, Luminal's $25M Series A and Databento's $100M Series C: all single source or aggregator relays, unverified.

28.

Menlo Ventures report data: single source (a 21st Century Business Herald write-up of a Science and Technology Daily reading of the report); the underlying report data is unverified.

29.

Zhipu's ARR of $1.6B: no new corroboration found in the window; still single source and unverified.

30.

T-Head's confidential A1 filing: single source (China Fund News via Xueqiu); the filing itself is unverified.

31.

The H3-Image license reading (not applicable in the US, EU, UK or South Korea; companies with over $1M in annual revenue need a separate written license): single-source interpretation, unverified.

32.

The Epoch AI dataset (top data centers at 15.1 GW of compute, 38 GW by end-2028): a single dataset, with power figures as estimates, unverified.

33.

Chainstack's AI-driven attack disclosure: single source; the incident itself falls outside the window, the disclosure inside it, unverified.

34.

Nebius' closing price: no reliable quote was obtained for Nov 12, described by its intraday decline instead; Nov 13 closed at $115.88 (-9.92%, Finnhub).

35.

Alibaba's Nov 12 morning-session fall of 4.97% to HK$95.6: reported only in a Chinese-language roundup, unverified.

36.

Zhang Yiming's remarks quoted verbatim: disclosed from an internal meeting, with no confirmation from ByteDance.

37.

Alibaba's claim that complex web tasks use half the tokens of DeepSeek 4.1 Flash: an official Alibaba claim appearing only in a Chinese-language roundup, unverified.

38.

Two brief incidents on OpenAI's status page on Nov 12: single-source aggregation, unverified.

39.

Anthropic's $10B-each private placement from Google and NVIDIA: partly confirmed — Fool and AI Agent Update reported the split from the filing on Nov 11, but no Reuters or Bloomberg cross-check; still unverified pending official or mainstream confirmation.

40.

DeepSeek Harness at 92 trillion tokens per day: company-reported, official-API-only figure with no methodology disclosed and no independent audit.

41.

Tencent Hy4 weights, system card and full-release license: Tencent only promised release "in the coming days" with no date; still not out as of Nov 13; the full-release license terms are unpublished, and the preview's Apache 2.0 license does not imply the full release's terms.

42.

Perplexity's new funding figures conflict: DefiLlama still lists a Nov 2 Series F-1 of $3B at a $30B valuation (updated Nov 6), but several Oct 31 checks say the last confirmed valuation was $20B in September 2025, with anything above $30B only in talks in August; no company announcement or mainstream confirmation — the $3B/$30B figures are not used.

43.

Claims that the Senate "passed the GAIN AI Act / AI chip export limits": the articles found carry inconsistent dates and contradict a Nov 12 Tikr piece saying the chip provisions did not make it into the House version; not usable as a new development in this window.

44.

TSMC's ADR close of about $406.27 (about -1.7%) on Nov 12 rests on a single MarketBeat table, unverified.

45.

Whether South Korea's PIPC has formally opened a case on Anthropic: Korean media on Nov 9 said pre-review, not yet filed, while TechTimes UK and AINewsToday on Nov 10 said formally filed, covering 112 accounts; the accounts conflict, with no new information in the window to clarify.

46.

Gemini 4 Carbon joining GitHub Copilot: announced Nov 10 (outside this window), with coverage still appearing Nov 13; priced at $2/$10 per million tokens, doubling from Dec 1. Not included in the main text; whether to add it is for the editor to decide.

47.

The gemini-2.5-pro API switching to Gemini 3.8 Flash responses from Dec 1: only OrcaRouter's reading of Google's deprecation page, with no official Google announcement cross-checked.