Transaction fraud: a 4.3x gap that narrowed
Stripe reports that attempted transaction fraud rates for startups on its network stayed roughly level over the past year, with AI companies as the clear exception. In Q3 2025, AI startups saw a 4.3x higher attempted transaction fraud rate than startups in aggregate. By Q1 2026, that multiple had fallen to 2.6x. The drop reflects faster adaptation to emerging fraud patterns by AI companies and by Radar; as transaction fraud becomes harder to pull off, attackers move toward other vectors such as multi-account and free trial abuse.
The underlying economics explain the targeting. AI products sell compute that is valuable and easy to resell. An attacker can buy a subscription or a block of tokens with a stolen card and then sell that access on secondary markets or in regions where the product isn't directly offered. The legitimate cardholder typically disputes the charge only after the resale is complete, leaving the business to absorb the loss.
Detecting this is hard for any single company because attackers rotate through stolen cards across many businesses. Radar draws on signals from billions of transactions across the Stripe network to flag fraudulent payments before processing, including cards already used fraudulently elsewhere even when they are new to a given merchant.
Sign-up abuse grew faster than payment fraud
Fraud is also increasingly showing up before any transaction occurs, which pushes companies to assess customer risk rather than just payment risk. In multi-account abuse, attackers register many accounts to repeatedly collect free tokens, trials and other new-account benefits. Stripe's models identified a 40% increase in attempted multi-account abuse across all AI subscription companies on its network between January 2026 and June 2026.
Among the most heavily targeted companies, the trend ran the other way from what stronger controls might suggest. The AI subscription companies with the highest attempted multi-account abuse rates recorded a 154% increase over the same six months, with some seeing increases above 600%. Stripe reads this as evidence that the most attractive targets are becoming more attractive still, and that attackers are shifting away from transaction fraud toward other forms of abuse.
Radar's abuse prevention features address this stage of the lifecycle by evaluating risk at registration and login. The multi-account signal flags one actor registering multiple accounts to exploit a service, while the account sharing signal catches a single account being used simultaneously from multiple locations.
What the data looks like in practice
Stripe points to its network scale—billions of transactions across millions of businesses—as the basis for detecting and blocking true fraud automatically. As one example, ElevenLabs used Radar's multi-account abuse prevention to block 2,000 users per day from abusing its free tier over a two-month period.
We also get so much fraud. Very common across most self-serve AI founders I speak with. I vastly underestimated how much fraud and abuse there is on the internet. Stripe has been incredibly helpful at blocking it for us.



