This article covers inference theft, where attackers steal paid AI API calls.
- •AI inference costs $2 per call, making theft millions of times more profitable than stealing HTTP requests
- •Attackers use residential proxies and adapters to present stolen APIs as OpenAI-compatible
- •Session authentication fails at scale; attackers bypass it using proxy IPs
- •Request-level verification is necessary to prevent profitable theft
- •Vercel's BotID uses invisible client-side ML to detect and block bot requests
This summary was automatically generated by AI based on the original article and may not be fully accurate.