Shops should be able to match public bills, seals and store videos to the originals; if a fake story spreads, collect evidence first, then report it to police or the platform (inference).
AI NEWS
Today’s AI News
Chinese and English are the same day. This page lists every item, each with the original link, the author, and the engagement.
Latest day
November 16, 2026
11-13 07:50 – 11-16 07:50 (UTC+8)
Top 3
This is a direction, not a same-day penalty rule; firms raising money or preparing AI-related disclosures can first check how they describe liability and data (inference).
If you have GPU or cloud deals in flight, watch the Nov 17 earnings call and existing orders rather than unofficial follow-on rumors (inference).
All
Shanghai Huangpu police say a viral “sky-high” Peace Hotel bill was an AI-altered receipt and staged video by a liquor seller; 15 people have been detained
Shops should be able to match public bills, seals and store videos to the originals; if a fake story spreads, collect evidence first, then report it to police or the platform (inference).
CollapseShanghai Huangpu police say a viral “sky-high” Peace Hotel bill was an AI-altered receipt and staged video by a liquor seller; 15 people have been detained
The Huangpu Public Security Bureau said on Nov 13 that a viral “sky-high drinks bill” from a well-known restaurant in the district was fabricated by the head of an internet liquor seller in another province, who sought attention and sales. Police said the group used AI to alter the bill amount, swapped in wines the company sells, produced a fake video, and pushed it widely through controlled accounts and paid posters. With help from police elsewhere, 15 people were detained; four were placed under criminal compulsory measures and 11 received administrative penalties. The official notice does not state the forged amount, so this item does not treat viral figures as official.
CSRC chair Wu Qing says China should deepen two-way AI and capital-market research, backing hard-tech firms while studying liability, data security and systemic risk
This is a direction, not a same-day penalty rule; firms raising money or preparing AI-related disclosures can first check how they describe liability and data (inference).
CollapseCSRC chair Wu Qing says China should deepen two-way AI and capital-market research, backing hard-tech firms while studying liability, data security and systemic risk
CSRC chair Wu Qing said on Nov 14 at the first China Capital Market Academic Annual Conference that China should deepen research on two-way empowerment between AI and the capital market. He said markets should strengthen rules on valuation and disclosure to support hard-tech firms, and should also study liability, data security, customer protection and systemic risk, while guarding against algorithmic black boxes and AI-enabled cyberattacks and hardening market infrastructure. The reports gave no timetable or quantitative targets. This item is based on published summaries and should not be read as a rule taking effect the same day.
NVIDIA on Nov 12: Jensen Huang resigns as president, CEO and director, effective immediately; the CFO becomes interim CEO, and fiscal Q3 results remain set for Nov 17
If you have GPU or cloud deals in flight, watch the Nov 17 earnings call and existing orders rather than unofficial follow-on rumors (inference).
CollapseNVIDIA on Nov 12: Jensen Huang resigns as president, CEO and director, effective immediately; the CFO becomes interim CEO, and fiscal Q3 results remain set for Nov 17
NVIDIA said on Nov 12 that co-founder Jensen Huang has resigned as president, chief executive and a director, effective immediately. CFO Colette Kress will serve as interim CEO and remain CFO until a permanent successor is named; the board said it will consider internal and external candidates and appointed Harvey C. Jones as its first-ever chair. Fiscal third-quarter results remain scheduled for about 1:20 p.m. PT on Nov 17. The announcement does not give a reason. This item was not carried in the Nov 13 edition and is added now from the official release, with the original date noted. Later reports whose full text was not reviewed are not repeated.
xFusion’s ChiNext IPO cleared review on Nov 12; media citing the filing say it plans to raise 8 billion yuan, the largest ChiNext listing filed in 2026, with registration still ahead
The 8 billion yuan is a planned raise, not cash in hand, and clearing review is not a listing; this is not a purchase or price-cut signal (fact).
CollapsexFusion’s ChiNext IPO cleared review on Nov 12; media citing the filing say it plans to raise 8 billion yuan, the largest ChiNext listing filed in 2026, with registration still ahead
The Paper reported on Nov 13 that the Shenzhen Stock Exchange listing committee, meeting on Nov 12, found xFusion met the conditions for a ChiNext IPO and for information disclosure; the exchange’s 61st review-result notice records the same conclusion. Citing the filing, the report said the company plans to raise 8 billion yuan, the largest ChiNext offering brought to review in 2026, for compute infrastructure, a smart manufacturing campus, critical technology research and working capital. Clearing the review is not a listing and the money has not been raised; registration, pricing and issuance still lie ahead. The fundraising totals come from media accounts of the filing; this item did not open the prospectus itself.
Shanghai Weibuild completes a fifth funding round of more than 100 million yuan and forecasts 2026 product orders above 100 million yuan, up more than 100%
Construction sites now have robot cases told in orders and job-site efficiency; the figures are a company forecast and interview claims, so rerun the math on your own site before buying (inference).
CollapseShanghai Weibuild completes a fifth funding round of more than 100 million yuan and forecasts 2026 product orders above 100 million yuan, up more than 100%
36Kr reported on Nov 15 that construction-robot firm Shanghai Weibuild (蔚建科技) had completed a fifth funding round of more than 100 million yuan, with Caitong Capital as the investor, mainly for manufacturing, overseas expansion and R&D. Citing founder Liang Yanxue, the report said the company expects 2026 product orders to exceed 100 million yuan, up more than 100% year on year, and that its plastering robot averages about 300 to 350 square meters a day. Those operating figures are a company forecast or interview claims. A headline line about a three-year compound order-growth rate above 100% was not independently laid out in the body and is not used here. The raise is described only as “more than 100 million yuan,” with no more precise figure.
China’s NDRC says the 15th Five-Year Plan will deepen an “AI+” campaign, starting with hazardous and labor-heavy scenes and expanding manufacturing, energy, transport and medical pilot bases
This is directional language for the 15th Five-Year Plan, with no budget figures, not a grant already awarded (fact).
CollapseChina’s NDRC says the 15th Five-Year Plan will deepen an “AI+” campaign, starting with hazardous and labor-heavy scenes and expanding manufacturing, energy, transport and medical pilot bases
The Cyberspace Administration of China on Nov 13 relayed a CCTV report from the National Development and Reform Commission: during the 15th Five-Year Plan period, China will let real-world use pull technical innovation and deepen an “AI+” campaign. Plans include projects that start with mining, high-altitude work, emergency response and other hazardous or labor-heavy settings, and continued national AI application pilot bases in manufacturing, energy, transport and healthcare, with faster standards and lower barriers to use. The report did not publish a budget or quantitative targets. The language overlaps an NDRC briefing on Nov 3; this item records the Nov 13 14:37 public report and does not treat it as a brand-new project list.
DeepSeek opens a full suite of app APIs: search and fetch at 5 yuan per 1,000 calls, file parsing and OCR at 1.5 yuan per 1,000 pages, with temporary 50% off speech recognition and TTS
Search, document reading and speech now have per-call or per-page prices; the 50% speech discount is temporary and should be recalculated at list price when it ends (inference).
CollapseDeepSeek opens a full suite of app APIs: search and fetch at 5 yuan per 1,000 calls, file parsing and OCR at 1.5 yuan per 1,000 pages, with temporary 50% off speech recognition and TTS
DeepSeek said on Nov 13 that it has opened a full suite of app APIs: with the right parameters on existing endpoints, V4.1-Flash (deepseek-flash) can do web search, URL reading and PDF/Office/image uploads; search, fetch, OCR, speech recognition and TTS are also available as standalone APIs. Official list prices are 5 yuan per 1,000 text or image searches and 5 yuan per 1,000 fetches, and 1.5 yuan per 1,000 pages for file parsing and OCR. Speech recognition is listed at 0.78 yuan per hour (0.39 yuan on a temporary 50% discount) and TTS at 3 yuan per 10,000 characters (1.5 yuan on the same discount); the pricing page does not give an end date. Tool fees are charged on top of model token fees.
Baidu AI Cloud’s standard Agent Memory service starts usage billing at midnight on Nov 16; compute and storage are not billed separately for now
If the preview was already wired into production, usage billing starts today—check the console. A unit price was not verified, so do not budget off second-hand figures (fact).
CollapseBaidu AI Cloud’s standard Agent Memory service starts usage billing at midnight on Nov 16; compute and storage are not billed separately for now
Baidu AI Cloud said its standard Agent Memory service would be formally released on Nov 16 Beijing time and start pay-as-you-go billing from midnight that day. Billing items are memory-reasoning input, memory-reasoning output and embeddings; compute and storage for the standard edition are not charged separately for now. The service had been free in public preview since May 29, 2026, and former preview users are billed from the same start time. The “billing overview” linked from the notice opened as VectorDB node prices, so this item does not invent a token unit price. The notice was updated on Nov 4; the effective date falls in this edition’s window.
Huawei Cloud launches openPangu-2.1-Pro at 0.8/3.2 yuan per million input/output tokens, available only in Southwest-Guiyang I
Teams already on Huawei Cloud or needing domestic compute can compare unit cost with the mid-tier model they use now; the service is currently limited to Southwest-Guiyang I (fact).
CollapseHuawei Cloud launches openPangu-2.1-Pro at 0.8/3.2 yuan per million input/output tokens, available only in Southwest-Guiyang I
Huawei Cloud updated its MaaS best-practice and text-generation pricing pages on Nov 15 to add openPangu-2.1-Pro. Huawei said the model improves on the previous generation in agents, code generation and throughput, and it is available for in-console chat and API calls. Official pricing is 0.016 yuan per million cached-input tokens, 0.8 yuan per million input tokens and 3.2 yuan per million output tokens, with no peak/off-peak split; the console price prevails. The service is currently limited to the Southwest-Guiyang I region. The official pages do not publish benchmark scores versus the prior generation.
Media reports say Ant’s Lingguang has passed 3 million hosted tasks and 11 million executions since 2.0, after earlier week-one figures that included 20 million usersUnconfirmed
These are “hosted task” counts, not downloads, and they come from media accounts of a company disclosure; treat them as a lead until an official source is available (inference).
CollapseMedia reports say Ant’s Lingguang has passed 3 million hosted tasks and 11 million executions since 2.0, after earlier week-one figures that included 20 million users
The Beijing News, Sina Tech and others reported that Ant Group’s AI app Lingguang on Nov 11 said hosted tasks had passed 3 million and cumulative executions had exceeded 11 million since version 2.0 launched on Oct 27. The same reports restated week-one figures disclosed on Nov 5, including more than 20 million users. Those numbers are media accounts of a company disclosure and were not independently confirmed on a Lingguang or Ant official blog, so this item is marked unverified and should not be read as an official announcement today. A Nov 13 LatePost piece was commentary, not a new data release. The figures date to Nov 11 and were not carried in the Nov 13 edition.
LangChain raises a $300 million Series C at a $3 billion valuation and says more than 7,000 organizations use LangSmith
Workday’s “five months to under a day” is a vendor case study; re-test on your own process before treating it as a timeline you will hit (inference).
CollapseLangChain raises a $300 million Series C at a $3 billion valuation and says more than 7,000 organizations use LangSmith
LangChain announced on Nov 13 a $300 million Series C at a $3 billion valuation, led by CapitalG, with participation from GIC, DFJ Growth, IVP and Sapphire Ventures. It said LangSmith, its platform to run and govern agents, is already used by more than 7,000 organizations. The company also said Workday cut the time to deploy a new agent from five months to under a day; that is a vendor case study, not a timeline others can assume. LangChain said it will show new LangSmith capabilities on Nov 20.
Anthropic launches Claude for Manufacturing: 17 skills to draft quotes and quality reports for a person to approve
It can draft quotes and quality reports for a person to approve; the connectors are mostly overseas software, and the page does not say how it will be sold or enabled in China (fact).
CollapseAnthropic launches Claude for Manufacturing: 17 skills to draft quotes and quality reports for a person to approve
Anthropic launched Claude for Manufacturing, a plugin for Team and Enterprise plans, with 17 skills and connectors to Autodesk Fusion, Infor, Microsoft 365 and Oracle NetSuite. It can draft job quotes, quality reports and supplier requests; a person reviews the work before it is sent, and Claude does not operate equipment. A technical webinar is set for 11:00 a.m. ET on Nov 17. The page does not say how the plugin will be sold or enabled in China, and the connectors are mainly overseas software. Trade press dated the launch to Nov 12; this is a catch-up item, and it is not the already-published Claude Meetings product.
In the Fed’s financial-stability survey, 87% of respondents flagged AI as a salient risk, up from 50% in the spring, citing valuations, cybersecurity and complex financing
This is a survey of worries, not a new ban; valuation, security and financing structure may come up more often in loan or fundraising talks (inference).
CollapseIn the Fed’s financial-stability survey, 87% of respondents flagged AI as a salient risk, up from 50% in the spring, citing valuations, cybersecurity and complex financing
The U.S. Federal Reserve on Nov 13 released its financial stability report. CNBC, citing the accompanying survey, said 87% of respondents named AI a “salient risk” to the financial system, up from 50% in the spring. The report said concerns centered on AI-related equity valuations, cybersecurity, and complex AI-related financing. This is a ranking of worries among market contacts, not a Fed finding that AI is already the largest systemic risk. The 87% and 50% figures were cross-checked against the report’s charts and CNBC’s account.
OpenAI launches ChatGPT Wallet: each payment needs biometric approval and the real card number is withheld; it starts with U.S. Pro users only
Agents are starting to pay for purchases, but this is only open to U.S. Pro users and does not help a domestic business yet (fact).
CollapseOpenAI launches ChatGPT Wallet: each payment needs biometric approval and the real card number is withheld; it starts with U.S. Pro users only
OpenAI announced ChatGPT Wallet on Nov 13, with American Express, Mastercard and Visa, so an agent can pay without seeing the customer’s real card number. Each purchase requires the user to review the item, total and card and approve with Face ID or a fingerprint; the real number is not shared with the merchant or the agent. The first release is through dots for U.S. ChatGPT Pro subscribers who are at least 18 and hold an eligible U.S.-issued card. The product is U.S.-only for now and is not available to businesses in China.
Anthropic opens a preview of self-hosted Claude Managed Agents: session history can stay in your cloud, but model calls still leave the network
Session history and credentials can stay on your side, but model calls still leave the network; pricing was not published, so smaller firms need not budget against it yet (inference).
CollapseAnthropic opens a preview of self-hosted Claude Managed Agents: session history can stay in your cloud, but model calls still leave the network
Anthropic said self-hosted Claude Managed Agents is in private preview: the control plane, credentials and session history can run in a customer’s own cloud or data center, but model requests still go to a chosen inference endpoint; the Claude model itself does not run locally. Docs require PostgreSQL 16 or later. The product page is undated; trade press and docs sites generally place it on Nov 12–13. Pricing was not published.
Fortune exclusive: OpenAI has disbanded its standalone robotics team, folding research and hardware into existing groups, which the company confirmed
OpenAI is folding robotics into its general-model teams, not announcing an exit; factory decisions should still wait on real-robot data and on-site deployment (inference).
CollapseFortune exclusive: OpenAI has disbanded its standalone robotics team, folding research and hardware into existing groups, which the company confirmed
Fortune reported exclusively on Nov 13 that OpenAI has disbanded its standalone robotics team. The company confirmed the change when asked, and said it was meant to move faster rather than to exit robotics. A spokesperson said research is joining the main research organization that reports to Mark Chen, and applied robotics and hardware are joining the applied organization that reports to Greg Brockman. The company argued that general-purpose frontier models can already handle some robotics tasks. There was no concurrent OpenAI blog post. Some financial figures in the article were not checked against primary statements and are not repeated here.
OpenAI launches Daybreak Red Team and Defense Factory to test web vulnerabilities and draft patches that engineers must review before they go live
Agents can find holes and draft patches, but an engineer still has to approve a merge; with no public price, do not budget as if they replace a whole security group (inference).
CollapseOpenAI launches Daybreak Red Team and Defense Factory to test web vulnerabilities and draft patches that engineers must review before they go live
OpenAI added two security tools under its Daybreak program; trade press generally dates the launch to Nov 12. Daybreak Red Team uses agents in an isolated environment to test web apps and confirm vulnerabilities; Defense Factory prepares patches, which engineers review and test before merge, and does not deploy automatically. No price was listed on the product pages. This is not the Nov 4 Daybreak launch on AWS GovCloud, which is older news.
Google lets AI Pro and Ultra subscribers claim Cloud credits inside AI Studio; credits last 12 months, and use requires a $5 prepaid balance
If the team already pays for Pro or Ultra, claim unused credits in AI Studio first; the page does not say whether a China-based account can complete billing (inference).
CollapseGoogle lets AI Pro and Ultra subscribers claim Cloud credits inside AI Studio; credits last 12 months, and use requires a $5 prepaid balance
Google said on Nov 13 that monthly Cloud credits included with AI Pro and Ultra subscriptions can now be claimed, linked to billing and tracked inside Google AI Studio. Each grant remains valid for 12 months and unused grants stack; only the primary subscriber can use them. Drawing on promotional credits for the Gemini API with Cloud billing requires a prepaid balance of at least $5. The post does not list the dollar amount of each plan’s monthly grant, so this item does not invent one. The Nov 13 news is the in-studio claim flow, not the first publication of the credit tiers. The page also does not spell out whether China-based accounts can complete billing.
ElevenLabs launches an experimental ElevenAI app that combines meeting notes, dictation and agent follow-up, currently in preview by signup
It ties note-taking to automatic follow-up, but is still a preview by signup; if you already use a notes tool, compare permissions and email hooks before joining the waitlist (inference).
CollapseElevenLabs launches an experimental ElevenAI app that combines meeting notes, dictation and agent follow-up, currently in preview by signup
ElevenLabs previewed ElevenAI and opened an early-bird signup at its New York summit on Nov 11; TechCrunch followed up on Nov 13. The experimental app puts meeting notes, dictation and agent follow-ups in one team product; dictation is said to cover some 90 languages, and it still requires a signup rather than a direct download. No list price was published, and the pages do not say whether it will work for users in China. This is a catch-up on a Nov 11 debut reported on Nov 13, not a new release today.
Mirendil announces a $1 billion Series A led by Kleiner Perkins and Clearlake Capital, without disclosing a valuation
This is a large lab financing, not a product you can shop against; the “under $10 million of compute” claim is the company’s own (fact).
CollapseMirendil announces a $1 billion Series A led by Kleiner Perkins and Clearlake Capital, without disclosing a valuation
Model lab Mirendil said on its site on Nov 12 that it had raised a $1 billion Series A led by Kleiner Perkins and Clearlake Capital, with participation from Google and others; no valuation was disclosed. The company said one researcher pre-trained its model with less than $10 million of compute in about 3.5 months. The announcement was not carried in the Nov 13 edition and is added now with the original date. It is distant from day-to-day decisions for smaller firms in China, so this item records the financing only and does not treat it as a product ready to buy.