Glacis

Glacis: AI Runtime Security and Assurance

Glacis is a runtime assurance infrastructure designed to secure AI systems by ensuring that every action taken by these systems is governed, verified, and auditable. Positioned inline at the AI request path, Glacis monitors actions, enforces policies, and signs receipts for every governed AI action without allowing sensitive data to leave the system. It functions as a proof layer that validates AI behavior in real time, catering to the needs of regulators, customers, and insurers.

"Policy defines. Guardrails enforce. Proof verifies." Glacis solidifies AI accountability by producing verifiable evidence for each action performed by AI systems. Its unique approach allows organizations to demonstrate compliance with regulatory standards while maintaining transparency with stakeholders.

Features

• Inline Monitoring: Observes every inference, tool call, and decision for comprehensive oversight.
• Policy Enforcement: Blocks unauthorized actions while allowing compliant requests to proceed.
• Verifiable Evidence: Generates signed and witnessed receipts for each action, ensuring independent verification without reliance on Glacis.
• Tamper-Proofing: Maintains a tamper-evident record of actions, enabling audits and compliance checks.
• Open Standard Support: Adopts the OVERT standard for runtime evidence, facilitating easy portability and accessibility of proof across different systems.

Glacis meets the heightened expectations for AI governance, particularly in industries like healthcare and finance, where accurate documentation of AI actions is paramount. It provides organizations with the tools to classify locally, enforce policies at runtime, and produce reliable proof of governance, all while enhancing security in AI operations.