Goods that travel by wire and are easy to resell. That is what an AI subscription looks like to a fraudster.
No need to steal a box and drive it across a border. You buy tokens with someone else's card, sell the access where the product is not available, and vanish before anyone has looked at their statement.
Fact
The facts: on 15 September 2026 Stripe published data from its network on fraud at AI companies. Per Stripe, in the third quarter of 2025 AI startups had a 4.3 times higher attempted transaction fraud rate than startups overall; by the first quarter of 2026 the gap had fallen to 2.6 times. At AI subscription companies, attempted multi-account abuse - creating piles of profiles for free tokens and trials - rose by 40 per cent between January and June 2026, and at the most affected by 154 per cent, at some by over 600 per cent. Stripe cites ElevenLabs, which over two months blocked 2,000 users a day from abusing its free tier. The post is by Jacob Meltzer, engineering lead of Stripe Radar, Stripe's fraud prevention product, and Samuel Bakouch, head of product for Payments Intelligence.
Read the numbers in order, because the order is the story. Attempted payment fraud falls, from 4.3 to 2.6 times. Attempted abuse at sign-up grows. The fraudster has not disappeared, he has simply moved from the till to the front door.
Lock the till and the thief goes to the door.
The free trial is the problem here. For ordinary software a trial is marketing and costs almost nothing. For an AI product it is compute that someone pays for, and a hundred fake accounts are a hundred electricity bills.
The post is, of course, also an ad for Radar. The numbers, though, come from a network that, per Stripe, sees billions of payments.
An AI product with a free plan is guarded at sign-up, not only at payment. By Stripe's reading, that is where the fraud attempts are moving.
Official primary sources
Stripe Blog - What Stripe data shows about fraud at AI startups, 15.09.2026The visual is generated code-art, no third-party images.