ChatPRD ends free trials after wave of token fraud
The AI product-spec tool's founder says fraudsters looped free trials and promo codes to resell model access and build Roblox and Fortnite maps.
By The Agentic Times ·

ChatPRD, an AI tool that helps product managers turn an idea into a written specification, is winding down its free trials entirely after sustained abuse by groups harvesting free model usage, according to an account published on 20 November 2026.
The details come from founder Claire Vo, who described the problem in a YouTube video and an accompanying post on Twitter, as summarised by developer and writer Simon Willison. Vo said she has seen a flood of what she calls token fraud: organised abuse of the product's free trial and promotional code system to obtain AI usage at her expense.
According to the account, the abuse was not limited to individuals wanting a few free queries. Vo said some of the activity was straightforwardly commercial. Attackers ran automated loops of free trials and promo codes, then "tunnelled" into her agent, using ChatPRD as a conduit to an underlying large language model for work that had nothing to do with writing product specifications. In one example she gave, she said she was effectively funding an international Roblox and Fortnite map studio, whose operators used her agent to build game maps.
Other abusers were reselling the tokens outright, Vo said, turning her free tier into a wholesale supply of model access.
One detail in the account points at how the abusers worked. Willison notes that one of the common prompts Vo saw running through her agent asked for an SVG image of a pelican riding a bicycle. He suggests the fraudsters were probably using it to identify which model sat behind the product, because different model families produce distinctive results for that prompt. He writes that he has used the same technique himself to guess the identity of stealth preview models on platforms such as OpenRouter. That interpretation is his, not a claim by Vo.
The case is a plain example of a cost-side risk that applies to any product built on top of a paid model API. The agent itself was not compromised in the usual security sense. There is no suggestion in the source that customer data was exposed or that the system was tricked into unsafe behaviour. The failure was commercial: a generous free tier, a promo code system that could be cycled automatically, and an agent interface flexible enough to be repurposed as a general-purpose model proxy.
ChatPRD has been live since the GPT-4 era, and Vo has built it in public for much of that time, according to the write-up. Ending free trials removes the obvious attack surface, but it also removes the standard way prospective customers try the product before paying. Willison describes the decision as sad but, in the face of the abuse, the right one.
The account does not say how much the abuse cost, how long it ran before detection, or how many fraudulent accounts were involved. Nor does it name the groups responsible beyond the description of game-map building work. No fix beyond withdrawing the free tier has been announced.
Sources
- simonwillison.nethttps://simonwillison.net/2026/Nov/20/token-fraud-is-here-and-its-bad
