Benchmarks, incident write-ups and engineering notes on securing AI agents in production.
To protect you, most AI security tools ask you to send them exactly what you are trying to protect. Why trust should be unnecessary, and why Cerbère AG is self-hosted by design.
91.5% recall at sub-millisecond latency from regex alone, 98.1% with a ML layer bolted on — and the false positives that come with it. Full methodology, mapped to the OWASP Top 10 for LLM Applications.
