The VEX Protocol is designed as a governed-execution and evidence architecture that can separate proposed action from authorized execution, revalidate state before consequence, and preserve cryptographically verifiable records for selected governed actions.
THE GOVERNANCE GAP
Autonomous agents can operate without formal boundaries or verifiable records. When an agent executes a tool call, teams may lack a standardized record of who authorized it, what goal it was pursuing, or what actually happened. Post-incident investigations can then be reduced to log analysis and speculation. VEX is designed to address part of this accountability gap by governing selected actions at the execution boundary.
PROTOCOL CONTEXT (GOVERNED EXECUTION)
The VEX Protocol is designed to support governed execution for high-risk agent actions in governed or pilot deployments. Its architecture separates proposal from authorization, validates state before consequence, and can preserve cryptographically verifiable evidence:
Proposal vs Authorization
Proposed actions are separated from authorized execution. The model proposes; the governance layer authorizes before consequence-bearing governed actions proceed.
State Revalidation
Execution state is revalidated before consequence. Pre-commit rebind ensures the target context matches what was authorized.
Fail-Closed Progression
If any validation fails, execution halts. The default is deny, not allow. Failed paths produce reviewable evidence.
Cryptographic Evidence
What crossed the boundary is preserved in tamper-evident form. The evidence is verifiable and auditable for compliance and incident response.
ProvnAI Mitigation
VEX is implemented through vex-core (the protocol library) and can integrate with McpVanguard (the enforcement proxy). Governed execution is intended to preserve cryptographically verifiable evidence at the execution boundary. Hardware-backed identity and TEE-backed deployment are being explored in pilot and research configurations as the stack matures.