Introducing Arcjet prompt injection detection. Catch hostile instructions before inference. Works with Next.js, Node.js, Flask, FastAPI, and any JavaScript / TypeScript or Python application.

Arcjet

Arcjet has launched AI prompt injection protection, a new security layer that detects and blocks hostile instructions before they reach an AI model. It targets jailbreaks, prompt-extraction attempts, shell injections, encoding attacks, and structural exploits like ChatML injection. The feature integrates into the application request lifecycle alongside existing Arcjet rules (bot detection, rate limiting, sensitive info detection). Code examples are provided for both Next.js with the Vercel AI SDK and Python FastAPI with LangChain. Pricing starts at $2 per 1 million tokens, and the feature adds approximately 100–200 ms of latency.

Introducing Arcjet AI prompt injection protection

Arcjet AI protection for production endpoints