Set hard spend caps and per-user rate limits before you launch any public AI bot or free trial that runs on a paid LLM API. Token fraud is now a real business, and organized groups use open AI agents as free compute.
Claire Vo, who runs ChatPRD, a tool that turns product ideas into specs, described a flood of free-trial abuse. Simon Willison wrote it up. Some abusers just wanted free AI. Others resold tokens or ran whole operations through her agent. One group cycled trials and promo codes to build Roblox and Fortnite maps on her bill. Claire is now ending free trials entirely.
What this means for a small business bot
A lead-intake bot on your website or a support assistant in Telegram is the same open door, just smaller. In my projects the defenses are simple. I set a monthly spend limit in the provider console and a daily token cap per user or phone number. I keep the system prompt narrow and max output short, so the bot is useless as a general-purpose model. I log cost per conversation and send an alert when a single chat spikes. These measures won't stop a determined group, but they make you a bad target and keep a bad night under a known number.
Why it matters: an AI bot without limits is an open tab on your credit card. Check your caps this week, before someone else finds the bot.
Source: Simon Willison