Shufti Ties AI Agent Actions to Verified Owners and Permissions

Identity verification firm Shufti has outlined an approach that ties each AI agent action to a verified owner and a defined set of permissions, Help Net Security reports. The proposal addresses one of the most practical obstacles to agent deployment: once software acts on a person's behalf, systems need to establish who it is working for and what it is allowed to do.

The report centres on two ideas, verified owners and permissions. An agent's behaviour is treated not merely as an API call but as an action attributable to a specific person or organisation and bounded by what that owner has authorised.

The framing follows a broader shift in the identity industry. KYC and anti-fraud tooling built for account opening, payments and login is being extended to machine identities, where the question is both who a party is and whether an agent is acting for them within granted scope.

For enterprises running agents, the appeal is accountability. When an agent moves money, places an order, queries a database or sends a message externally, a dispute requires a clear chain showing who authorised the action, how far the authorisation went and whether the action stayed inside it.

The hard part is granularity. Agent behaviour is often multi-step and decomposed by the model itself, so a permission model designed around a single call can break down midway through a chain. Binding identity and permissions at the action level means carrying tokens, context and audit records through every hop.

Public details so far remain at the level of positioning; product packaging, integration paths and customer references are not described in the report summary. What can be confirmed is a direction: identity governance for agents is moving from an accessory feature to a standalone infrastructure question.

What to watch next is how such schemes interoperate with existing IAM and OAuth stacks, and whether they hold up under high-frequency, long-chain agent calls. If identity and permissions can genuinely travel with each action, the barrier to putting agents into finance, support and internal workflows drops considerably.