Establish responsibility
Connect the agent to its operator and the person or organisation granting authority.
Know Your Agent
Give AI agents
clear authority.
Know who is behind an agent, what it may do, and why an action was allowed.
Legis Labs is developing Agent Passport to connect delegated authority, checks before execution and evidence for review in regulated workflows.
Design partner enquiries welcome
LEGIS LABS / AGENT PASSPORT
Authority, made
inspectable.
inspectable.
ACCOUNTABLE OWNERExample Treasury Ltd
AGENTpayments-assistant-07
DELEGATED SCOPEApproved supplier payments
AUTONOMOUS LIMIT£1,000 per action
PROPOSED ACTIONPay supplier · £1,200
Human approvalAuthority checkedPolicy appliedDecision recorded
Illustrative concept · synthetic data · no live transaction
For the people accountable
when AI takes action.
Financial institutionsInsurersProfessional servicesAI developers
The question behind every action
Identity is the start.
Authority needs evidence.
A recognised agent can still exceed its mandate. Useful controls connect identity to the specific action, the applicable rules and the person who is accountable.
Check the proposed action
Apply explicit permissions, limits and approval requirements at the point where an action can be enforced.
Preserve the evidence
Record the mandate, policy version, decision and outcome so another reviewer can examine what happened.
Our work
Verification at the points
that matter.
Agent authority is our priority. Our work in AI assessment and client due diligence brings the same discipline to evidence and review.
01 / KYA02 / ASSURANCE03 / ONBOARDING
Agent Passport
A proposed framework for bounded permissions, action checks and an inspectable audit record.
Development & pilot programmeAI & agent assessments
Scoped assessments of AI systems and agent workflows, grounded in evidence, repeatable tests and clear limitations.
Discuss an assessmentKYC & AML due diligence
Client onboarding that connects identity, document and multimodal biometric evidence to professional review.
Scope and methods agreed per engagementThe assurance principle
Check the control.
Show the limits.
A deterministic rule can test a defined action against a defined mandate. The quality of the evidence and the coverage of the control still matter.
We design for explicit outcomes: allow within scope, block outside it, or pause for a named approver. Assessments should show which paths were tested, where uncertainty remains and what would trigger a fresh review.
Explore our approach to assurance ↗C Combinator · a community initiative by Legis Labs
Bring the ecosystem
into the conversation.
Practical AI safety needs people who build systems, use them, regulate them and study their consequences. C Combinator creates room for that exchange.
Explore C Combinator ↗ResearchPracticePolicyCollaboration.
Research & perspectives
A field moving quickly.
A reason to be precise.
What should an agent passport actually prove?
Identity, delegated authority and evidence have different jobs.
Read the note ↗What makes an AI audit useful to a buyer?
Scope, reproducible tests, conflicts and limitations.
Read the note ↗The emerging KYA ecosystem
A source-linked introduction to identity, payments and runtime controls.
Explore the field ↗Work with Legis Labs
Put a real workflow to the test.
Bring an agent use case, an accountable owner and a decision you need evidence to support.