BriefTechNews
OpenAI's ChatGPT Wallet Lets Agents Spend Your Money - With Biometric Approval

OpenAI's ChatGPT Wallet Lets Agents Spend Your Money - With Biometric Approval

7 min read · 13 sources

TL;DR
  • OpenAI launched ChatGPT Wallet with biometric-per-transaction approvals for agentic purchases, built with Amex, Mastercard, and Visa.
  • Anthropic launched five enterprise products ahead of an IPO expected Nov. 19 or 20 at a potential $1.4 trillion valuation.
  • Kalshi is in talks to raise funds at a $65 billion valuation for a GPU and cloud compute futures platform.
  • An FBI-discovered North Korean operation called FaceStealer stole more than $400 million in crypto over 18 months using AI video impersonations.
  • Jane Street more than doubled its Nvidia stake in Q3, holding 23.8 million shares worth about $5.44 billion.

The future of payments is an agent spending your money while you hold the keys. OpenAI’s ChatGPT Wallet launches today for US Pro users through Dots, letting AI agents complete purchases without ever seeing your actual card details - but every single transaction needs your biometric approval. It’s the first serious answer to the question every fintech engineer has been asking for a year: how do you let software spend money without turning your card into a firehose?

It’s not alone. Anthropic dropped five enterprise products in one day ahead of an IPO that could value the company at $1.4 trillion, Kalshi is raising at a $65 billion valuation to build futures markets for GPU compute, and the FBI just exposed a North Korean operation that stole $400 million in crypto using AI-generated faces on video calls. The agentic economy is arriving - and so are its attack vectors.

Anthropic’s revenue grew 1,206% to $33.2 billion through the first nine months of the year, ahead of an IPO that could value it at $1.4 trillion.

ChatGPT Wallet: Payment Tokens, Not Card Numbers

Source: openai.com ↗

ChatGPT Wallet is built with American Express, Mastercard, and Visa, and it uses payment tokens that preserve existing card rewards and dispute protections. The architecture matters: the agent gets a token that can only be used for a specific transaction amount at a specific merchant, not a reusable card number. Each purchase requires biometric approval from the user, which means the agent can’t run up a tab while you’re asleep.

For engineers, the interesting part is the fraud layer. OpenAI is working with payment processors and fraud providers to help merchants recognize legitimate agent-driven transactions. That’s a real problem: when a machine is buying things, the behavioral signals that fraud systems rely on - typing speed, device fingerprints, session patterns - all collapse. Merchants need a way to say “this is an approved agent, not a bot attack.”

Anthropic's Five-Product Blitz Ahead of a $1.4 Trillion IPO

Source: finance.yahoo.com ↗

Anthropic launched Claude Meetings, Claude Whiteboards, Claude Sheets, Claude Data, and interactive connectors for third-party software - all in one day. Claude Meetings joins video calls, follows conversations, and answers questions in the meeting chat. Claude Whiteboards turns planning discussions into timelines with assigned responsibilities. Claude Sheets builds spreadsheets from working data. Claude Data aggregates information from multiple platforms into dashboards, summaries, or videos.

The IPO is expected Nov. 19 or 20, with a potential valuation up to $1.4 trillion. Revenue grew 1,206% to $33.2 billion through the first nine months of the year. For anyone running SaaS infrastructure, the signal is unambiguous: Anthropic is coming for the seats Salesforce, ServiceNow, and Intuit occupy. Their shares fell earlier this year on similar news, and this launch is the confirmation.

Kalshi Wants to Hedge Your GPU Bills

Source: reuters.com ↗

Reuters reported on November 12 that Kalshi is in talks to raise fresh funding at a $65 billion valuation for a platform where companies can hedge against changes in the cost of renting GPU or cloud computing capacity. This moves Kalshi beyond event contracts - the prediction-market-style bets it became famous for - and into a market that CME is also pursuing.

The use case is real. If you’re running a training cluster, your largest variable cost is compute, and it swings wildly. A futures market for GPU rental prices lets you lock in costs the way airlines hedge fuel. The engineering challenge is defining the underlying index: what exactly is “the price of a GPU”? Kalshi will have to solve the same problem CME is working on, and the first one to a credible benchmark wins.

Stripe's Tempo Bet: Payments-First Blockchain

Source: paymentsdive.com ↗

Stripe-backed Tempo is positioning itself as blockchain infrastructure built specifically for payments. The key design choice: reserving most block space for simple money movement so low-margin transactions aren’t crowded out by higher-value trading activity. That’s the opposite of general-purpose chains like Ethereum, where a single DeFi trade can spike gas prices and price out a $2 cross-border payment.

Tempo is targeting cross-border payments, payroll, marketplace payouts, and emerging agentic commerce. Stripe’s Bridge and Privy acquisitions provide the stablecoin and wallet infrastructure around the chain. The bet is that stablecoin payments need a chain optimized for settlement throughput, not smart contract flexibility.

Revolut's AI Compliance Shortcut

Source: youtube.com ↗

Nik Storonsky explained in a 52-minute interview how Revolut plans to challenge Amex and Chase in the US, and how the company uses retention and gross profit benchmarks from past launches to decide which new products to scale. The most striking operational detail: Revolut is using AI to shorten compliance reviews that have taken 210 days.

That number matters for anyone in fintech. Compliance reviews are the bottleneck in every financial product launch, and a 210-day cycle means you can’t iterate. If Revolut has genuinely compressed that to days, it has a structural advantage over every incumbent that still runs manual KYC and AML reviews.

Stock Tokens Are Where Stablecoins Were in 2023

Source: fintechbrainfood.com ↗

Tokenized stocks and ETFs total about $3.2 billion in distributed tokens - roughly 0.004% of the market - against about $431.6 billion in real-world assets onchain per RWA.xyz. The field splits into four models, from wrapped tokens with no shareholder rights to natively issued stock, and 56 of 62 companies opted out of the OKX and ICE pooled offering.

The more significant development is the DTCC’s Digital Tokenization Service (DTS), which received less attention than the OKXICE snub. The DTCC is the plumbing of the US securities market; its DTS is a concrete infrastructure step for tokenized assets. The refusals from Apple, Nvidia, Microsoft, and the rest of the Magnificent Seven highlight persistent trust and regulatory hurdles.

Google Wallet Meets Gemini

Source: blog.google ↗

Google is rolling out a connection between Google Wallet and Gemini’s Personal Intelligence feature for all US Gemini app users. Users can tag @Google Wallet to ask questions about spending, budgeting, and subscriptions. Gemini can generate dynamic charts, flag recurring charges and price increases, and suggest holiday budgets. Users can optionally link financial accounts via Plaid for deeper insights.

The critical constraint for engineers: Gemini is read-only. It can analyze transaction data but cannot move, manage, or invest money. The integration is opt-in, and the new Memory Summary feature provides a readable snapshot of saved preferences for user control. That’s the right architecture for an AI with access to financial data - read access for analysis, no write access for action.

Wells Fargo's AI Teammate Is Working

Source: bankingdive.com ↗

Wells Fargo’s AI Teammate has logged hundreds of thousands of interactions since launching last quarter, according to wealth and investment management AI head Andre Mansour. The tool runs as an agent over Advisor Gateway’s 200-plus adviser tools and cuts meeting prep from hours to a couple of prompts.

The forward-looking claim: Mansour expects advisers to become an approval layer for clients’ own AI agents. That’s the same pattern as ChatGPT Wallet - the human approves, the agent executes. It’s becoming the standard architecture for AI in finance.

FaceStealer: $400 Million in Crypto, Stolen With Fake Faces

Source: thefinanser.com ↗

The FBI discovered a North Korean cybercrime operation called FaceStealer that stole more than $400 million in cryptocurrency over 18 months. The criminals used AI-generated impersonations of known colleagues on video calls, tricking victims into downloading malware that steals credentials, accesses crypto wallets, and captures webcam footage. The FBI recovered over 900 webcam recordings, showing victims becoming unwitting faces for further attacks.

The attack vector is human trust, not technical weakness. The same playbook could hit banks - urgent client transfers, account recovery with a matching face and voice. For engineers, the lesson is blunt: biometrics and video confirmation can no longer stand alone. Verify identities through out-of-band channels.

Quick Hits

Source: paymentsdive.com ↗

Western Union terminated its $500 million deal for Intermex, saying conditions set by California’s DFPI were unacceptable. Barclays is preparing to cut about 200 investment banking roles in New York and London as part of its annual performance review, on top of roughly 5,000 job cuts announced last month. PayPal wants an industrial bank charter to support merchant lending and diversify funding. And Jane Street more than doubled its Nvidia stake in Q3 to 23.8 million shares worth about $5.44 billion - though the 13F filing omits short positions and options notional values, so don’t read it as a directional bet.

Get the brief

Liked this one? Pick the beats you care about and get just those, five minutes a morning. Free, no hype, one click to leave.

About Me Author

My name is

BriefTechNews

A daily digest of what actually moved in AI, tech, crypto and fintech, assembled and written with AI, with every item linked to its source. Read More
Tags

You May Also Like