EU AI Act β Article 14
EU AI Act Article 14: evidence of human oversight
Article 14 requires high-risk AI systems to be designed so that natural persons can effectively oversee them β including the ability to intervene in or halt an operation. INTYGA holds an AI agentβs tool call until a person signs its exact parameters, so that oversight leaves a receipt you can show to someone who does not trust you. Article 14 does not require cryptographic proof, and this alone does not establish conformity.
First, the correction
What evidence to keep for each overseen action
Keep records a reviewer can check without trusting you:
- The signed approval: the person's signature over the exact action and its parameters, with the nonce and expiry it was bound to.
- Who approved, kept separate from the system that requested the action, and how their credential was enrolled.
- The tamper-evident record of each decision, including refusals and requests that expired unanswered β an intervention that stopped an action is oversight too.
- Which actions no person approved. Anything a policy approved automatically is marked as such, carries no human signature, and should not be presented as oversight.
- Where it matters, evidence that the approved action took effect, from a source independent of the system that ran it.
Producing the first and the last of these is covered step by step in how to prove a human approved an AI agent's action and how to verify an AI agent's action actually succeeded.
What Article 14 asks for, and what the control leaves behind
The tool call blocks at requireApproval until a human approves it, denies it, or it times out. That is a circuit breaker in the execution path, not an alert raised after the fact β and the distinction is the one Article 14 is about, since oversight that arrives after the transfer has settled is not oversight.
The approval is a signature over the exact parameters the agent proposed. Verification re-derives those bytes from what is about to execute, so the record shows the approver saw this instruction and not a summary of it. One byte of drift fails verification.
The approving human's identity is bound into the signature and kept distinct from the machine that requested the action. requesterType: AI_AGENT with approverType: HUMAN is explicit in the record, and so is the case where policy auto-approved and no person was involved.
Enforcement sits in the gateway the agent connects through, not in a prompt asking the model to behave. Policy is controlled by your organization and encrypted such that INTYGA cannot read it.
Why a log is the weaker answer
Most oversight evidence today is an application log written by the system under review. The party that signed the evidence is the party you would be auditing, which is exactly the position an assessor is trained to distrust.
A client-signed receipt moves the trust anchor. The approver's authenticator produces the signature, the relying party verifies it against parameters it derived itself, and neither step requires an INTYGA secret or an INTYGA API call.
Both halves are open specifications an assessor can read: DIV defines the signed approval and its offline verification, and DEWP defines the tamper-evident record of each decision.
const action = {
target: BILLING_TARGET, // this service β so the approval can't be replayed against another
actionType: "issue_refund",
params: { customerId, amountCents: 480000 }
};
const approval = await intyga.requireApproval("Refund β 4,800.00 EUR", action);
if (approval.status !== "APPROVED") throw new Error("not approved");
// nonce and approvers come from YOUR side β a receipt cannot vouch for its own signer.
const anchor = parseTrustAnchorFile(readFileSync(process.env.INTYGA_APPROVERS_FILE!, "utf8"));
const TRUSTED_APPROVERS = trustAnchorApprovers(anchor); // quorum counts PEOPLE, not keys
const check = verifyApprovalReceipt(approval.receipt!, {
tenantId: process.env.INTYGA_TENANT_ID!, // YOUR trusted workspace configuration
...action, nonce: approval.nonce!, approvers: TRUSTED_APPROVERS,
}, {
expectedOrigin: anchor.webauthn?.origin,
expectedRpId: anchor.webauthn?.rpId,
});
if (!check.ok) throw new Error("receipt mismatch: " + check.reason);
await issueRefund(customerId, 480000);ISO/IEC 42001 A.9.2 β the standard that operationalizes it
Article 14 states an obligation; ISO/IEC 42001 is the AI management system standard an auditor will actually work from, and its Annex A.9.2 governs human oversight directly. The same three controls carry both.
High-risk tool calls hold at require_approval until a human signs or the challenge expires.
Canonical parameter rendering puts the exact JSON the agent proposed in front of the approver, on their own device.
Requester and approver are separate identities by construction; four-eyes policy forbids self-approval.
Start with the agent that worries you most.
You do not need a governance programme to begin. Gate one destructive tool call, verify the receipt yourself, and decide from there whether the evidence is worth having on the rest.
See the other framework mappings β