Tag
This paper presents Canonical Action Verification and Attestation (CAVA), a runtime-semantics layer that converts heterogeneous agent activity into canonical runtime action objects for governance, enabling consistent approval binding, receipt reproducibility, and policy enforcement across diverse AI execution environments.
The article discusses the concept of verifiable AI inference, exploring methods like trusted attestation and cryptographic proofs to ensure the authenticity and provenance of AI-generated outputs without rerunning the model.
The article explains how a single XSS vulnerability can defeat the phishing-resistance of passkeys when attestation is set to 'none', allowing attackers to register their own passkeys and achieve persistent account takeover. It calls for attention to this overlooked threat and suggests defenses.
A deterministic action-level attestation architecture for AI mediation was developed and validated in discussions with Microsoft's engineering team. The author seeks investors or partners for the software architecture.