Legis Labs research · updated 26 October 2026
Earn trust through
inspectable work.
Notes on agent authority, independent assessment and the emerging KYA ecosystem. We distinguish standards, vendor statements and our own proposals.
Research note 01 · Legis Labs analysis
What should an agent passport actually prove?
A useful passport connects an agent to an accountable principal and a current, specific mandate. Identity answers who or what is presenting a credential. Authority answers whether this proposed action is permitted in this context. Evidence lets another party examine the decision later.
A signed credential alone does not enforce a permission. The tool or service receiving a request needs to validate the issuer, agent binding, scope, expiry and revocation, then apply its own policy. The action that executes must match the action that was approved.
Standards and industry initiatives already address parts of this problem. Google’s Agent Payments Protocol describes payment mandates, while the Monetary Authority of Singapore’s SAFR initiative proposes governance checkpoints before agent actions execute. Legis Labs’ proposed contribution is a practical evaluation of how controls and evidence work together in a bounded regulated workflow.
Sources: Agent Payments Protocol · MAS SAFR, 3 July 2026
Research note 02 · Legis Labs analysis
What makes an AI audit useful to a buyer?
An audit should answer an agreed question about an identified system. It needs a versioned scope, evidence of the work performed and conclusions that stay within what the tests establish. A score without the method and failure cases is difficult to use in procurement or risk review.
For an agent workflow, an assessment can test permission boundaries, approval integrity, revocation, evidence completeness and the response to adversarial inputs. The report should distinguish a measured failure from an untested risk, and describe when changes to a model, tool or policy require a retest.
Independence also needs a definition. Being separate from the LLM provider does not automatically make an assessor independent of the whole system. A provider assessing its own enforcement layer should disclose that relationship and involve an external reviewer where an independent conclusion is required.
Sources: NIST AI Risk Management Framework · FCA AI practice review, October 2026
Field notes · 26 October 2026
KYA is becoming an ecosystem.
Identity providers, access-control vendors, payment networks and runtime-security teams approach agent trust from different starting points. Product pages describe intended capabilities; they do not establish independent effectiveness or adoption.
Identity and credentials
Trulioo currently labels its KYA product Alpha and describes a Digital Agent Profile. Sumsub announced a KYA module for MCP on 19 October 2026.
Action controls
APort and AstraSync publish approaches covering agent mandates and checks around actions. Their claims and maturity need assessment in the context of a specific deployment.
Payments and interoperability
AP2, Visa’s Trusted Agent Protocol and related initiatives address identity, intent and payment interaction. Institutions retain their own policies and integration requirements.
Our working view: buyers should compare the complete path from principal verification to enforcement and independently reviewable evidence. A familiar label or a passport format alone is insufficient.
Selected reading · not endorsements
Follow the original sources.
These sources inform the wider field. They do not imply that their authors or institutions endorse Legis Labs.
Work with Legis Labs
Contribute evidence to the discussion.
We welcome reproducible evaluations, integration experience and well-defined problems from across the ecosystem.
