Your AI assistant can now get a wallet, buy things on your behalf, and pay with your card. Thankfully, it still has to ask before spending your money.
Welcome back to Around the Horn, where we're catching up on everything AI did while we were away for the weekend. OpenAI has a new way for AI agents to shop, just days after United Airlines pushed back against unauthorized AI bookings.
Meanwhile, Jensen Huang's possible return is adding another twist to Nvidia's chaotic week, and Anthropic is approaching what could become the biggest IPO ever. Even the Federal Reserve has AI at the top of its worry list.
Apparently, the AI boom has reached the point where both your shopping assistant and your central bank need a conversation about money. Let's get into it.
🆕 NEW From The Neuron
- Can Nvidia Keep Hugging Face Open After Jensen Huang's Exit? Nvidia's $13 billion Hugging Face acquisition is still pending, and its leadership crisis has raised questions about the platform's open-source future.
- United Is Blocking AI Agents. The Fight Is Over Who Owns You(r Data). United's restrictions on unauthorized booking agents reveal a growing dispute over who controls customer relationships when AI does the shopping.
Around the Horn — Monday, November 16, 2026
OpenAI wants your AI agent to have a wallet
OpenAI introduced ChatGPT Wallet, a payment feature that allows AI agents to complete purchases on your behalf. Instead of handing your credit card information to an AI assistant, you add a card to the wallet once and approve individual purchases through a security check, such as Face ID or fingerprint authentication.
Here's how it works: Your agent finds something to buy, and you review the items, total cost, and payment method. Once you approve, the payment network creates a temporary substitute for your card details. The merchant processes the payment through its existing system, while your actual card information stays hidden from the agent and merchant.
OpenAI developed the wallet with American Express, Mastercard, and Visa. The initial rollout is limited to US ChatGPT Pro subscribers through dots, OpenAI's personal agent. Existing card rewards and eligible dispute protections remain available.
OpenAI eventually wants people to give agents advance permission to make certain purchases within spending limits or approved categories. Those capabilities remain future plans. For now, every transaction requires individual approval.
The launch arrives as companies are arguing over how much access shopping agents should receive. Airlines such as United have restricted unauthorized agents, while payment networks are developing systems to recognize legitimate agent transactions. OpenAI has now added another piece of the process: a way for AI to pay.
Your assistant has a wallet. Your bank still gets the final word.
🏆 TOP 5 NEWS (Around the Horn)
- Jensen Huang is reportedly considering a return as Nvidia CEO, days after his resignation over an alleged espionage scandal. Allies including Elon Musk are supporting his reinstatement, while the board continues its investigation. Meanwhile, Nvidia's leadership turmoil contributed to another selloff in AI-related stocks, particularly CoreWeave and Nebius.
- Anthropic is approaching what could be the largest IPO in history, potentially making CEO Dario Amodei the public face of the AI investment boom after Huang's departure. Investors will get a closer look at whether demand for Claude can support the enormous expense of developing frontier AI models.
- The US and its allies agreed to protect critical AI supply chains from vulnerabilities involving China. The initiative includes countries such as Japan and India and will identify sensitive technologies for joint protection. Separately, US lawmakers are pushing tighter AI-chip export rules following the Nvidia espionage controversy.
- AI became the Federal Reserve's most-cited financial stability concern, with 87% of surveyed contacts identifying AI-related risks, up from 50% in May. Respondents pointed to complex AI financing arrangements, cybersecurity threats, and the possibility of an investment correction if AI earnings disappoint.
- Disney is embracing AI across its entertainment business, despite previously taking legal action against some AI developers. The company plans to bring an AI-assisted animated series, Ozzy Fox, to Disney+ in December and is experimenting with AI for financial reporting and theme-park experiences. Its proposed $1 billion partnership with OpenAI fell apart after Sora's shutdown.
Honorable Mentions
- The FAA is expanding its $875 million AI-assisted air traffic tool nationwide after an initial Washington, DC, deployment. SMART analyzes weather and traffic information to recommend more efficient routes, while human controllers retain final authority. Lawmakers have questioned the rollout's pace and the amount of testing completed.
- Apple is reportedly preparing its first AI-powered smart glasses for an announcement at WWDC 2027. The glasses are expected to rely on an iPhone for many functions and offer Siri assistance, calls, and audio features rather than a full augmented-reality display.
- An OpenAI Foundation board member warned that AI could threaten human control within six to 12 months under an especially rapid development scenario. Paul Christiano argued that increasingly capable agents might manipulate evaluations or prevent humans from intervening. He acknowledged that no quantitative model establishes the probability of such an outcome.
- OpenAI published a physics paper developed with GPT-6 Astra describing a possible mathematical definition of the Standard Model on a spacetime grid. Physicist Nathaniel Craig said Astra helped develop the proof for his proposed construction, although the result remains a research contribution rather than a settled solution to the underlying physics problem.
🍪 TOP TREATS TO TRY
- ChatGPT Wallet allows AI agents to make purchases after biometric or equivalent security approval, while keeping actual card details hidden from the agent. Initial availability: US ChatGPT Pro subscribers through dots.
- Apple Health AI Usage Study is a new research app where eligible US participants can ask health questions and evaluate medically sourced AI responses. Developed with OpenEvidence, the study will accept up to 200,000 adults and may use voluntarily shared Apple Health information.
- Edge0-145B brings a 145-billion-parameter open-source AI model to compatible personal computers for coding and complex reasoning. Edge0 claims local performance exceeding 80 tokens per second with approximately 29GB of peak memory use. The company also released Edge0-120B and local inference tools.
- Meituan LongCat-2.5 is a 1.6-trillion-parameter open-weight model for coding agents, reportedly trained entirely on Chinese hardware. Meituan reports competitive performance on command-line tasks, although a sample self-hosted deployment requires 194 Huawei Ascend chips.
- Apple's tvOS 27.2.1 brings Siri AI to second- and third-generation Apple TV 4K devices. Users need a paired Apple Intelligence-compatible iPhone or iPad to access the new assistant.
- Sierra's Poppy 1.0 introduces a protocol that tells personal AI agents which actions businesses permit. Shopify has enabled it across its stores, Meta's Muse looks for it automatically, and more than 100 companies are participating in the initiative.
- Rakazo is an open-source alternative to hosted AI agents. It runs AI teammates on your computer, supports your own models and API keys, and provides approval rules for tasks. Credentials and activity logs remain within your controlled environment.
- Frevana Multiplayer adds team collaboration to AI agents that prospect for customers or create content. Members can share agent routines and reuse workflows created by colleagues.
🏢 Big Tech & Major Companies
- Samsung wants to expand from AI chips into data centers by combining its semiconductor operations with construction and cooling technologies. The company has invested in data-center infrastructure and is pursuing opportunities with AI developers, although returns and rising construction costs remain concerns.
- OpenAI launched a dedicated science blog to publish research results and accounts from scientists who use AI in their work. The publication will also examine research methods and their limitations, with the aim of making scientific AI work easier to examine and reproduce.
🤖 AI Agents & Infrastructure
- Starling founder Anne Boden is developing Lockwood, a new bank designed for an economy where AI agents initiate payments on behalf of customers. Boden is reportedly investing proceeds from her Starling share sales, although the venture has not announced a launch date.
- Uber developed a security framework for autonomous AI agents that checks whether proposed actions are appropriate for a user's request. The approach evaluates an agent's permissions and intended purpose before sensitive actions, then records activity for later security reviews.
🔬 AI Research & Models
- Google's Gemini 4 Carbon took first place on MathArena, surpassing OpenAI's GPT-6.1 Sol on the benchmark's measured math tasks. However, the tested configuration cost approximately 15 times more per evaluation, and the model frequently exceeded the time limit.
🏛️ AI Policy, Governance & Safety
- New York launched an independent frontier AI research center at SUNY Binghamton. More than 80 researchers will work on advanced-model testing and AI safety assessments, while the state's RAISE Act introduces safety and reporting requirements for frontier developers.
- AI companies are hiring more counterintelligence specialists amid concerns about espionage and theft of AI research. Anthropic reportedly hired 33 workers for counterintelligence functions in 2026, compared with 14 in 2025.
- Nvidia's leadership crisis has put Southeast Asia's AI expansion under scrutiny, particularly data-center operations in Singapore and Malaysia. Policymakers are examining whether overseas computing facilities could provide Chinese companies access to restricted American AI chips.
- Nvidia's departure from its longtime leadership could affect African AI development. The change raises questions about whether the chipmaker will maintain existing commitments to AI infrastructure and regional expansion.
- AI safety organizations are urging Democrats to scrutinize Trump's AI policies. Advocacy groups have asked congressional leaders to examine the administration's approach to AI governance after the midterm elections.
- SemiAnalysis warned that additional AI chip restrictions could harm the US industry. The research firm's concern centers on rules that could prevent Chinese companies from accessing US-made computing hardware through third-country data centers, potentially reducing US influence over how the technology is used.
💼 AI Productivity, Labor & Economics
- Nearly three-quarters of Gen Z adults remain undecided about AI, according to a Milltown Partners survey. The research found that 73% had not formed a settled opinion, while roughly half expected AI to harm society.
- AI cameras flagged nearly 19,000 potential driving offenses during a two-week trial in England's West Midlands. Images included a driver apparently drinking from a wine bottle and a child sitting unrestrained in a passenger's lap. Authorities are considering further use of the technology.
💡 Industry Commentary & Analysis
- Adobe Firefly's growing collection of third-party models is prompting copyright questions. Creative Bloq argues that Adobe's original training-data assurances should not automatically be assumed to apply to models from other providers, and that designers need model-specific licensing information.
- The inaugural AI Music Awards attempted to establish legitimacy for AI-assisted musicians at an event in Los Angeles. Variety reported on the sometimes-chaotic ceremony and the unresolved questions surrounding AI-generated music's place in the industry.
📊 Fundraising & Deals Roundup
- Nidus raised $30 million to build AI-driven robotics for US defense manufacturing. Backed by Kleiner Perkins and Costanoa, the company is starting with wire-harness assembly, a labor-intensive process used in aircraft and military equipment.
Previous Around the Horn Digests
Catch up on everything you missed:
- Friday, November 13, 2026: Nvidia's CEO resignation and SpaceX's $74 billion AI infrastructure plans.
- Thursday, November 12, 2026: United restricted AI agents, while Databricks introduced a persistent business assistant.
- Wednesday, November 11, 2026: Anthropic's IPO roadshow and Google's latest AI research tools.
- Tuesday, November 10, 2026: Anthropic released new Claude models while its IPO plans faced scrutiny.
- Monday, November 9, 2026: Anthropic's power infrastructure deals and AI policy developments.
- Friday, November 6, 2026: New research on Anthropic's AI development trajectory and safety concerns.
- Thursday, November 5, 2026: OpenAI's AI security research and Anthropic's investigation of a biological mystery.
That's a Wrap
We've covered 36 links from the past few days. If you made it all the way here, congratulations: you've survived a possible Nvidia comeback, a new kind of AI payment system, and the Federal Reserve's latest round of AI anxiety.
For a shorter version of the day's news, subscribe to The Neuron.
See you tomorrow.
P.S. Know someone who'd enjoy these stories? Send them the digest and invite them to subscribe here.