2026-11-20 · ← News
Token fraud forced ChatPRD to close its free trials
Token abuse cost ChatPRD tens of thousands of dollars, according to founder Claire Vo, and led her to end free trials. The case shows how a valid AI SaaS account can become someone else's compute infrastructure.
The image could not be loaded.
ChatPRD founder Claire Vo says token fraud cost her company tens of thousands of dollars over the past year. Abusers cycled through free trials, resold promotion codes and routed work for other services through the product. The company is now ending free trials.
A free account started funding someone else's game studio
ChatPRD sells an AI product for product managers with seat based pricing. Vo says one abusive operation routed work through her agent for a studio making Roblox and Fortnite maps. Others wrapped the product in relay services and resold model access.
Vo describes six patterns: trial farming, promotion abuse, card fraud, relays and resale, misuse of custom MCP connections and model probing. One custom MCP acted as a persistent compute worker. A valid login could not show whether a customer was using the product or another business was using it as infrastructure.
AI SaaS sells compute even when the invoice lists seats
For conventional SaaS, an active user may add little marginal cost. In an AI product, every call to an expensive model creates a direct bill. Seat based pricing built around average use breaks when someone turns a trial into a cheap inference endpoint.
ChatPRD therefore tracks dollars rather than request counts alone. It uses a model allowlist, spending limits, Stripe Radar, automated traffic checks and host restrictions for custom MCPs. Suspected abuse can be moved to a cheaper model. That is more useful than one message limit because two requests can have radically different costs.
A harder gate blocks fraudsters and some good customers
Ending trials also affects people who honestly wanted to test the product and companies that might later have paid. Vo also describes blocking entire countries and acknowledges losing legitimate users there. Fraud defense is a business tradeoff, not a clean technical optimization. It shifts cost from the operator to customers.
The automated classifier has no published evals or false positive rate. Vo says ChatPRD corrected one false positive during a large ban wave. Appeals and bounded support permissions matter as much as detection itself.
Margin will reveal the attack before a security dashboard does
AI companies should watch costs by account, organization, model and access path in near real time. A sudden rise in spend without matching revenue or product activity is the important signal.
The next evidence will come from other founders and providers of payment and AI gateways. If relays, trial farming and MCP abuse recur, free trials for expensive agent products will shift from a growth tool to a luxury small companies cannot afford.
Lilith's verdict
ChatPRD paid the bill for a studio building Fortnite maps through its agent. When a trial opens a back entrance to expensive compute, a growth funnel quickly becomes someone else's power socket.