GLACIS Technologies is building a sensor between an app and a model to verify the controls promised are executing and write a tamper-evident record.

AI systems are no longer just answering questions. They are calling tools, handling sensitive data, making recommendations and taking consequential actions inside real businesses. The question is not only what the model can do. It is whether the rules meant to keep it in bounds actually governed the action — and whether anyone outside the software stack can prove it. Glacis is the runtime supervision and evidence layer for AI that acts. At the point where an AI system is about to act, Glacis applies approved rules; allows, blocks, redacts, escalates or requires human review; and creates a signed operational record another party can independently verify. The proof can travel to customers, auditors, regulators, insurers and boards while protected prompts, responses and source data remain in the customer’s environment. It records which control ran, which policy version applied, what decision it made and whether the record still verifies. Glacis works with the governance platforms, guardrails, policy engines and observability tools teams already use. We add the missing link between policy, control decision and proof: supervision for AI you operate, evidence for AI you sell, and independently checkable records from AI vendors you oversee. We created OVERT, an open, royalty-free standard for portable runtime evidence, and placed its stewardship outside the company so it can outlast any one vendor. AI needs supervision. Prove yours had it.

Founded 2025