CTGT

CTGT is an applied interpretability lab that develops a policy engine constraining frontier AI models to complex organizational policies at the moment of output, ensuring reliability and compliance for mission-critical applications. The company productizes mechanistic interpretability, making it possible to edit the behavior of large language models to add safety policy guarantees without retraining, delivering determinism far beyond simple prompting. CTGT's Policy Engine acts as a logic compiler that dramatically improves accuracy in regulated workflows, demonstrated by doubling legal reasoning performance from 39% to 78% where standard RAG pipelines failed, and maintaining 96% integrity in entity resolution tasks. The platform serves financial services with SEC, FINRA, and organizational compliance applied instantly, insurance with 30% boosts in policy adherence, and achieves up to 80% lower inference costs by pairing open-source models with governed policy constraints. Founded in 2023 and headquartered in San Francisco, the company has raised approximately $7.7M in seed funding from investors including Google's Gradient and General Catalyst. CTGT differentiates itself through deterministic policy graphs governing agent outputs at inference time with cryptographically attestable audit trails, validated by industry leaders including the creator of Keras.