Key Takeaways
- Free tiers attract users who burn high-cost frontier models on tasks completely unrelated to your core product.
- Promo code abuse often operates as organized resale rings on platforms like Reddit to spin up separate businesses on your bill.
- Attackers use reverse proxies and automated agent loops to treat AI SaaS platforms as backend compute engines, including generating Roblox and Fortnite maps.
- Custom MCP integrations open a dangerous vector where attackers connect rogue servers programmed to query models continuously without sleep.
Free Trials Become Free Compute for Strangers
When Claire Vo built ChatPRD, she expected product managers to draft requirements documents. Instead, she spent a year fighting attackers who treated her application as an unmetered proxy for expensive models. The bill ran into tens of thousands of dollars.
The earliest leaks came through basic product generosity. Vo found that offering free access without aggressive controls created an open invitation for compute arbitrage. Fraudsters signed up for free tiers, ignored the product's actual purpose, and routed external tasks to the most expensive LLMs available.
“What we see is a lot of customers, you know, I'm going to call them fraudsters,” (play from 6:07) Vo said. “A lot of fraudsters signing up for free tiers and then really using the most expensive models to do non kind of like product aligned work.” (play from 6:12)
Attackers bypassed payment gates by cycling through batches of stolen or prepaid cards. Even paying tiers leaked cash when lifetime limits and velocity caps were missing.
Promo Codes Fuel Resale Rings and Automated Studios
Promo codes turned out to be an even bigger leak than open signups. Vo discovered that marketing codes meant for conferences or communities were scraped, bundled, and traded openly online.
Once bad actors secured codes, they automated the extraction. One ring hijacked ChatPRD to power an external production pipeline. “I am funding a international Roblox and Fortnite map studio,” (play from 11:34) Vo explained. “And so they like run a loop of free trials and promo codes and then tunnel to my agent and use it to build Roblox maps, I guess, Fortnite maps.” (play from 11:40)
The attack profile shifted from individual freeloaders to commercial operations using reverse proxies to hide behind regular user traffic.
Custom MCP Servers Turn Features into Compute Drains
AI platforms increasingly allow users to bring custom tools, but open extensibility creates direct infrastructure exposure. ChatPRD allowed custom Model Context Protocol connections so product teams could link internal company data. Attackers exploited that capability to run automated, indefinite loops.
“Somebody built like a monster MCP and then like installed that as a custom MCP on Chat Purity because we let you install custom MCPs because we want you to have your business data connected,” (play from 9:59) Vo said. “And then it became this like 24/7 worker that was prompting and saying like, I'm never sleeping.” (play from 10:09)
Instead of querying a company database, the malicious server turned ChatPRD into a captive worker engine, looping requests around the clock. If you give users the ability to connect persistent external endpoints, you are effectively giving them control of your API keys unless you place hard fences around request duration and token consumption.
What to Do With This
Audit your LLM application tomorrow morning for single-user velocity caps and promo code rate limits. Restrict your most expensive models to paid accounts that have cleared anti-fraud checks, and block prepaid cards on high-cost routes. If you support custom MCP servers or external tools, enforce strict timeout windows and kill loops that run without direct user interaction in the active browser session.