How Stripe Is Cracking Down on Token Thieves Bleeding AI Startups Dry

Patrick Collison didn't sugarcoat the problem. One in every six new sign-ups at AI companies using Stripe now involves token theft. Fraud has doubled in six months, and the pace accelerated sharply last month. The CEO laid out the numbers on a recent podcast: crooks create fake accounts to snatch tokens that buy computing power, then resell them or burn them on their own schemes. AI startups foot the bill. Free trials, once a simple growth tool, have become a liability. Some companies have dropped them entirely.

Emily Sands, Stripe's head of data and AI, calls this one of the most overlooked issues in artificial intelligence. Attackers burn inference costs at machine speed, racking up massive bills they never plan to pay. The scam is like diners who eat and run, but it happens in seconds. Fraudsters sign up across multiple AI services, deploy automated agents, and consume tokens in minutes. Traditional fraud teams can't react fast enough. A $500 token pack meant to hook new users vanishes. The perpetrator disappears. Growth stalls.

Stripe has moved aggressively. At its Sessions 2026 conference in late April, the company unveiled 288 new products and features targeting this pain. Expanded Radar now scans sign-ups and usage in real time, pulling signals from across Stripe's network. It assigns risk scores based on IP addresses, email domains, device fingerprints, and more. Results came quickly: for eight high-growth AI businesses, Radar blocked over 3.3 million risky sign-ups in a single recent month. Conversion rates for legitimate customers jumped from one in 25 to one in three when fraud dropped. ElevenLabs and Lovable already rely on the updated tool.

Stripe is also building infrastructure for real-time payments. Streaming payments let companies charge for every token the instant it's used, combining usage tracking from its Metronome acquisition with micropayments on the Tempo blockchain via stablecoins. The idea: when tokens flow like money, payments must match that speed. Traditional billing cycles can't keep up with machine-speed consumption. The new model reduces abuse because there's no large upfront credit to steal. Payment happens continuously. Risk stays low.

The company hasn't stopped at fraud and payments. Agent wallets, powered by its Link consumer wallet (over 250 million users), let people authorize AI agents to make purchases on their behalf. A one-time card is issued for each task. Real payment details stay hidden. Approval remains with the human. Stripe also partnered with Google to let businesses sell directly inside AI Mode and the Gemini app, with similar deals following from OpenAI, Microsoft, and Meta.

Analysts have taken notice. TD Cowen's Bryan Bergin called Sessions Stripe's largest product launch event, with over 9,000 attendees watching live demos of Meta, Google, and Shopify running agentic commerce in production. The company's valuation climbed to $159 billion in a February tender offer, a 74 percent increase from the prior year. Payment volumes keep climbing as the AI wave builds.

Sands put the stakes plainly: bad actors no longer limit themselves to stealing money or credentials. They steal tokens, each with real cost. The problem has become urgent for any business embedding AI. Stripe's bet is that the company that powered payments for the internet can also power the economics of intelligence. The thieves have had their run. The infrastructure layer is fighting back.

Source: Webpronews