Verifiable AI

Verifiable AI refers to systems designed so that their outputs, training data, or behavior can be independently checked by a third party rather than taken on the developer's word. Common approaches include cryptographic proofs that a model ran as claimed, audit logs that record how a decision was reached, and third-party testing against defined standards. For compliance and risk teams, this matters because regulators and customers increasingly expect evidence, not just assurances, that an AI system behaves as described.

1 item

Corporate PolicyGlobal2026-10-19

Anthropic Acquires Opaque Systems, Bringing Verifiable AI Controls In-House

Anthropic has acquired Opaque Systems, a startup specializing in techniques that protect sensitive data during AI processing and allow independent verification of AI behavior. Opaque's published tools include an Agent Governance Toolkit and a tracing framework called TRACE. It also published Confidential MCP, a protocol for securing agent tool connections, and a Weight Custody Manifest for model provenance. The acquisition shifts capabilities that enterprise compliance teams have been evaluating as third-party controls directly into Anthropic's product and infrastructure stack.

Anthropic Opaque Systems verifiable AI agent governance vendor concentration risk confidential computing AI auditability