A KYC vendor has built the KYA chain into a product: prove the agent, prove the human who authorised this specific action, enforce what that human allowed, and keep a record that survives review.
What happened
Shufti, a London-based identity verification vendor, announced Know Your Agent (KYA) on 28 October 2026 — an agent verification flow aimed at businesses receiving requests from autonomous agents that open accounts, place orders or move funds for their owners.
The pitch is that a valid access token cannot answer the three questions a relying party actually has. In the company’s words: “A valid access token alone cannot answer these questions. Stolen credentials can appear legitimate, and agent impersonation allows fake agents to pose as trusted ones.”
The product’s moving parts, as described in the release:
- Agent credential. Each agent receives a unique, time-bound credential that can be revoked. Known agents are recognised by name; unfamiliar agents are assessed through behavioural biometrics, and each agent carries a trust score.
- Principal binding. Every agent is linked to an individual or business verified through Shufti’s existing KYC or KYB checks. The company is explicit that KYA extends those checks rather than replacing them.
- Mandate. The principal sets a mandate per agent covering permitted actions, limits, expiry and revocation, enforced as a policy on every request, with monitoring for misuse.
- Step-up authorisation. Requests outside the mandate, or above a risk threshold, trigger a human-in-the-loop liveness check — a selfie — before proceeding. Shufti says its single-selfie liveness is iBeta Level 3 certified.
- Audit trail. Requests, decisions and revocations are recorded against both the agent and its principal.
Shufti says it already recognises consumer agents including Meta Muse, GrokBot and OpenAI Dots, “within a week of their launch,” and gives examples of policy shapes: a bank allowing balance checks and approved billers while requiring further checks for new payees; a crypto platform allowing trading within a limit while blocking withdrawals.
Chief Technology Officer Faryam Asif framed it as an accountability record: “Shufti Know Your Agent verifies every AI agent request with a verified owner, a clear mandate, and a record that holds up under review.”
Why it matters for KYA
This is the first time the whole KYA chain has shipped as one commercial flow. Agent identity, principal binding, scoped mandate, step-up consent and an audit trail have been discussed separately — in NIST’s design tenets, in IETF delegation drafts, in bank agent policies. Shufti has packaged them as a single decision point at the moment a request arrives. That is the shape a relying party actually needs, and it is a useful reference architecture even for teams that build rather than buy.
Name recognition is not attestation, and the gap is the interesting part. Known agents are “recognised by name” and unknown ones are scored behaviourally. That stops a casual impersonator, but it is reputation and fingerprinting, not cryptography: an attacker who can present as Muse gets evaluated as Muse. A signed agent credential bound to a registered principal would settle that question differently — and the standards work to produce one (WIMSE workload identity, SCITT execution evidence, delegated-authority chains) is still drafts. Shufti’s own time-bound revocable credential is the bridge: it makes revocation real today, while its trustworthiness still rests on Shufti’s assertion rather than a verifiable chain.
“Mandate” is doing the load-bearing work. The release’s examples are all limits and exclusions — approved billers only, no withdrawals, customer-defined budgets. That is delegated authority expressed as policy, which is what a KYA check has to produce: not a yes/no on the agent, but a decision on this request against this principal’s stated scope. A trust score alone cannot do that.
The audit trail is the deliverable an auditor will ask for. Linking every request and revocation to both agent and principal is what turns “we verified the agent” into evidence. The same record is what a bank, a payments scheme or a regulator will demand after an agent-driven loss, and it is the part most agent deployments still lack.
Caveats
Everything here is Shufti’s own description of its product; no independent evaluation, published specification, pricing or customer deployment has been released, and the iBeta certification covers the liveness check, not the KYA decision. The list of supported agents (Muse, GrokBot, Dots) is a vendor claim about coverage, not a register, and Shufti is not an authority that certifies those agents’ identities. Trust scores and behavioural biometrics are proprietary and unreviewed, so their error rates in blocking legitimate agents — the failure mode that costs a business revenue — are unknown. Nothing in the announcement says how a business would verify Shufti’s assertions if it did not trust Shufti.
Sources
- Shufti, “Shufti Launches KYA to Verify AI Agents and Human Authorization”, press release, London, 28 October 2026 — shuftipro.com.
- Shufti, “What is Know Your Agent (KYA)?”, knowledge base — shuftipro.com.
- Shufti, AI agent verification product page — shuftipro.com.
Know Your Agent (KYA) explains agent identity, verification and accountability. This is an explainer, not legal or compliance advice — see our Sources & methodology. New to KYA? Start with What is Know Your Agent? and the glossary.