Fast & Accurate Prompt Injection Detection API

ZooClaw introduced a security API to block prompt injection in AI agent workflows.

Most requests are handled by a fast DeBERTa-based first-stage classifier, while uncertain or risky inputs are escalated to a larger LLM.

The system uses a fail-closed design so ambiguous cases are handled conservatively.

The company says benchmark results outperform competing systems while keeping latency low.