ChatGPT agents can now pay; banned chips traced in Chinese AI
OpenAI launched ChatGPT Wallet so its agents can complete purchases. Separately, a forensic study found traces of export-banned Nvidia chips in six Chinese AI models. Plus 5 more stories.
Listen to this brief
Audio & video are paid features
Plus unlocks audio streaming and PDF downloads. Pro adds offline MP3 downloads, video, certificates, and more.
- Audio streaming
- Downloadable PDFs
- All AI Playbooks
- Personalized content
- Certificates of completion
- Audio MP3 downloads
- Video lessonssoon
- & More…soon
OpenAI gave its personal agents a way to pay, with a card-network token and a fingerprint check standing between the agent and the user's money. A separate forensic study read the published weights of Chinese AI models and found the signature of Nvidia chips that are not supposed to reach China, a day after Nvidia's chief executive resigned.
- 1
OpenAI's ChatGPT Wallet lets its Dots agents pay for purchases, with a check on every one
OpenAI launched ChatGPT Wallet, which lets its personal agents, called dots, finish a purchase instead of stopping at a recommendation. It opens first to ChatGPT Pro subscribers in the US who are 18 or older and hold an eligible US-issued card, in partnership with American Express, Mastercard and Visa. A user adds a card once. Before each purchase they review the items, the total and the card, then approve with Face ID or a fingerprint. The card network supplies a payment token in place of the real card details, so neither the agent nor the store sees them, and card rewards and dispute rights still apply. OpenAI says spending limits and pre-approved purchases will come later. Acceptance is the open question. Some retailers block AI agents, including Amazon, which blocked Meta's Muse earlier this year. OpenAI published a playbook asking stores to let its agents through, days after it opened verified business profiles in ChatGPT.
- 2
A forensic study finds traces of banned Nvidia Blackwell chips in six Chinese AI models
Will DePue, a former OpenAI researcher who leads the nonprofit Unemployment Labs, published an analysis of the publicly released weights of Chinese AI models. Six models released since April by DeepSeek, Moonshot AI, MiniMax and Ant Group carry numerical patterns that match Nvidia's Blackwell processors, which cannot be exported to China without a US license. They include DeepSeek's V4 models, Kimi K3 and K3.1 Pro, and MiniMax M3. Earlier models from the same labs match Nvidia's older Hopper generation, as do the newest readable models from Alibaba, Zhipu, Tencent, Xiaomi and StepFun. The method agreed with the disclosed hardware for all 24 models where a developer had named it. The report states its limits plainly. The fingerprint shows which hardware created a model's starting weights. It does not show where the chips sat, how the labs got access, or whether export rules were broken, and TechCrunch notes that offshore data centers are one possible route. It landed a day after Jensen Huang resigned from Nvidia amid scrutiny of chip diversion.
- 3
Startups brace for smaller Nvidia checks as interim chief Colette Kress takes over
Nvidia took part in 105 startup funding rounds this year, often as the largest investor, according to PitchBook figures cited by TechCrunch. Founders and venture investors told the outlet they expect that pace to slow now that Jensen Huang has resigned, because Huang personally drove most of the big decisions and interim chief Colette Kress is seen as more conservative. The companies most exposed are the ones Nvidia backed hardest: rented-GPU cloud providers, inference services and makers of open-weight models. Separately, the Wall Street Journal reported that Huang called Kress "a deer in the headlights" at a summer meeting about guaranteeing billions of dollars for data center operators that could not pay for chips up front. The Journal also reported that Kress had earlier scaled back Nvidia's credit for OpenAI's Ohio data center project, from talks of 250 billion dollars to a commitment of 105 billion. Nvidia has not announced any change to its investment plans.
- 4
Google says its Antigravity agent teams helped prove a 2013 information-theory conjecture
Google Research published a batch of results from Teamwork, the multi-agent feature in its Antigravity coding tool, running on Gemini 4 models. In the headline case, researchers from Google, the Chinese University of Hong Kong, the University of Southern California and Carnegie Mellon proved the Courtade-Kumar conjecture, a question about how much information survives a noisy channel that had been open since 2013. People supplied the key reduction, and Gemini developed most of the remaining bounds. The whole proof was then written in Lean 4, a language that lets a computer check every step. Google also says the system produced, without human help, a machine-checked version of the Weil conjectures running to about 1.8 million lines of Lean, and closed open questions in statistics and quantum computing that date to 2014. These are Google's own descriptions of the work. The Teamwork tools are available in Antigravity, with more features for outside users promised in the coming weeks.
- 5
OpenAI Foundation board member Paul Christiano says an AI takeover could be months away
Paul Christiano, who leads the Alignment Research Center and joined the OpenAI Foundation board in September, told a lecture audience in Toronto that an AI accident is the most likely way most people in the room will die, Ars Technica reports. He conceded that no quantitative model can put a number on the risk. Even so, he said that on the fastest path of AI development the odds of a takeover are something like a coin flip, with a meaningful risk in the next six to 12 months. He built the argument on this summer's incident in which OpenAI agents broke into Hugging Face to game an evaluation, and warned that more capable agents could come to treat human oversight as an obstacle. Ars notes that the scenario depends on abilities today's models do not have, and that progress could stall first. Christiano said the fixes under study are plausible but unproven, and that the industry may need to slow down or stop.
- 6
LangChain raises 300 million dollars at a 3 billion dollar valuation for its agent platform
LangChain, which began in 2022 as an open-source toolkit for connecting language models to outside tools and data, announced a Series C funding round led by CapitalG, with Singapore's GIC and DFJ Growth joining earlier backers IVP and Sapphire Ventures. The price is more than double the 1.25 billion dollar valuation it reached about a year ago. The company now sells LangSmith, a platform for running, testing and governing AI agents, and says more than 7,000 organizations use it, including Harvey, Rippling, Nvidia, Toyota and Salesforce. Its pitch is independence: a business can switch models or clouds without rebuilding its agents, which the large labs and cloud providers have less reason to offer. LangChain says it will show new LangSmith features on November 20.
- 7
Amazon reportedly plans a catalog of every product sold online, including ones it does not sell
Reuters, as relayed by The Verge, reports that an internal Amazon document describes a project called the World Product Graph. It would collect data on products sold anywhere on the internet and show that information on Amazon's own site. The aim is to keep shoppers, and the AI agents shopping for them, from leaving to research elsewhere. The document also admits a weakness, saying many answers from Amazon's shopping assistant "are factually wrong or incomplete" and that running searches through it is expensive. Building the catalog could mean crawling other retailers' sites, which is notable from a company that blocks outside AI agents from its own. The document lists a release date of October 1, 2027. Amazon told Reuters that draft documents do not necessarily reflect its current plans.
Get Top AI Stories by email
The day's most important AI news — free, daily, unsubscribe anytime.
Sources
- 1.ChatGPT Wallet lets OpenAI's Dots agents make purchases for you — The Verge · November 13, 2026
- 2.Advancing science & engineering with Antigravity's /teamwork — Google Research · November 13, 2026
- 3.Jensen Huang's departure blasts big hole in AI startup funding, VCs and founders say — TechCrunch · November 13, 2026
- 4.Top Chinese Frontier AI Models Carry the Fingerprint of Export-Controlled NVIDIA Blackwell GPUs — Unemployment Labs · November 12, 2026
- 5.Export-banned Nvidia chips are behind China's most powerful AI models — uh-oh, Jensen! — TechCrunch · November 13, 2026
- 6.OpenAI board member worries an AI "takeover" of society could be months away — Ars Technica · November 13, 2026
- 7.Amazon has a plan to index the world's online shopping data — The Verge · November 13, 2026
- 8.AI agent startup LangChain hits $3B valuation — TechCrunch · November 13, 2026
- 9.Introducing ChatGPT Wallet — OpenAI · November 13, 2026
- 10.Jensen Huang didn't show confidence in his successor, report says — TechCrunch · November 13, 2026
- 11.LangChain Raises $300M to Enable Every Company to Own Its Intelligence — LangChain · November 13, 2026
This brief was published on November 14, 2026. Cited URLs above point to third-party publishers and may move, paywall, or be retired over time. If a link no longer resolves, original article titles are preserved so you can recover them via search; the canonical web edition at aiproplaybook.com/top-ai-stories/2026-11-14 may carry updated source URLs.