About Compiled

We built the immune system
we wished existed.

Compiled is built by security engineers and machine learning researchers who got tired of watching governance fail at machine speed.

Our mission

Compiled mathematics. Not more rules.

Enterprise AI is eating the world. Governance infrastructure hasn't kept up. We believe the answer isn't more rules or more prompts — it's compiled mathematics. A 7KB function that can't be jailbroken, can't be prompted around, and runs in under 100ms.

The immune system metaphor isn't marketing. It's architecture. Your body doesn't run a rule lookup when it encounters a pathogen — it runs a compiled response. We built the same thing for enterprise AI.

0.995 AUC

Detection accuracy on production detectors

<100ms

Scan latency at p99

~7KB

Per compiled antibody

The team

Built by people who've been in the room.

CEO

Co-founder & CEO

Former CISO at a Fortune 500 financial institution. Led security through 2 regulatory examinations and an AI governance programme from inception.

CTO

Co-founder & CTO

ML research background. Former applied science lead at an enterprise security company. Published work on semantic embedding and anomaly detection.

VP Engineering

Security engineering background. Former principal engineer at a cloud infrastructure company. Specialises in zero-trust architecture and regulated environments.

Advisors

Expertise where it matters.

Enterprise Security Advisor

Former CISO at two publicly-traded financial institutions. Expert in regulatory examination preparation and enterprise security architecture.

ML & AI Advisor

Professor of machine learning at a top-10 research university. Published work on differential privacy, federated learning, and semantic embeddings.

Regulatory Advisor

Former senior examiner at a financial regulator. Expert in SEC, FINRA, and FCA communications surveillance requirements.

What we believe

Our values.

Precision over vagueness

We publish AUC numbers, false positive rates, and re-identification probabilities. Not marketing claims.

Transparency about limitations

Synthetic-trained detectors have a scope boundary. We tell you exactly where it is and how to calibrate.

Customer data stays with customers

Zero-egress isn't a feature. It's the architecture. We can't see your data even if we wanted to.

Speed without shortcuts

Sub-100ms isn't an accident. We removed the LLM from the inference path because it had to go, not because it was easy.

Want to learn more?

We're happy to walk through the architecture with your security team.