AI Token Theft Surge: How Fraudsters Steal Billions in Free Trials and How Startups Fight Back

Patrick Collison, the chief executive officer of payment processor Stripe, has warned that a surge in token theft is impacting the AI economy, with fraudulent actors now accounting for one in every six new customer signups at AI firms.

According to Collison, cybercriminals are defrauding AI companies by creating multiple new accounts to steal tokens used to purchase computing power. These tokens are then either resold or utilized for criminal activities.

Speaking on the TBPN podcast, Collison stated that token pilfering has become so prevalent that it is becoming prohibitively expensive for AI startups to provide free trials to potential customers.

While traditional software trials typically incur minimal costs for the provider, AI services require significant compute resources. Emily Sands, Stripe’s Head of Data and AI, noted that many startups provide token packs valued at $500 in an effort to convert trial users into paying customers.

This economic structure creates a unique opportunity for theft compared to traditional software, as the tokens provided during signup have tangible resale value.

Sands described the speed and scale of these attacks as a primary challenge for the industry.

I think token theft is the most under-discussed topic in AI. One of the things that’s really scary about that is that these attackers can burn inference costs, can rack up massive usage bills that they never intend to pay, and they can do that very, very quickly because they are consuming tokens at machine speed.

Emily Sands, Stripe’s Head of Data and AI

In an interview with Fortune, Sands explained that fraud typically involves individuals signing up for multiple accounts across various AI companies. These actors use the tokens for purposes unrelated to the firm’s intended service or resell them before disappearing.

Sands compared the scam to people who dine and dash at restaurants. The problem is further exacerbated by the use of AI agents, which allow attackers to exhaust tokens within minutes. This speed often prevents anti-fraud managers from intervening in time to stop the loss.

The epidemic has forced some startups to eliminate free trials entirely, though this strategy can hinder growth by limiting the acquisition of new users.

To address these losses, Stripe is adapting its Radar fraud detection tool for its credit card payment network to identify and block token fraud. The system assigns real-time risk scores to new accounts by analyzing device fingerprints, email domains, and IP addresses.

Companies including the app builder Lovable and the voice detection service ElevenLabs are currently using the service to block attempts at multi-account signups.

Sands reported that Stripe data indicates startups can improve their customer conversion rates from 1 in 25 to 1 in 3 by blocking these fraudulent accounts.

Beyond fraud detection, new payment infrastructures are being developed to mitigate the risk of token theft. One such approach is streaming payments, where customers use stablecoins to pay for AI services in real time as they are consumed.

Stripe is supporting a new blockchain called Tempo to enable these payments, while Coinbase has developed a corresponding product named x402.

Despite these developments, the frequency of fraud continues to rise. Sands noted that free trial abuse more than doubled between November 2025 and May 2026, with a dramatic acceleration occurring in April 2026.