Give a person the final say over your AI.
And have the EU AI Act evidence ready.
On the enterprise plan, register each AI system, assess it and link its controls to the approvals, runs and judges your agents already record. Export a report your auditors and counsel can read.
Hiring with AI, with a record of every safeguard.
One company uses AI to answer candidates and draft shortlists. Each safeguard leaves a record in connic, and the export pulls them together.
The careers chat tells each candidate that an AI assistant answers. The company's site records that notice in connic for every conversation.
A candidate says the rejection ignored six years of payroll work. The team opens an incident linked to the screening run and adds a corrective action with an owner and a due date.
Pick the system, environment and quarter. connic freezes the record and exports a readable report, without prompts or model outputs.
Oversight controls point to the decisions your team made.
connic turns the approved assessment into controls, each with its article, owner and status. Where your agents already record evidence, such as recruiter approvals or traced runs, the control links to it.
Marking a control not applicable needs a written rationale.
See early which obligations may apply.
Answer questions on EU scope, your role and how the system is used. connic derives a preliminary result and keeps every version with the reviewer's reasoning.
- Every change is a new version
- New answers create a new version. Approving it needs a written reason.
- Preliminary by design
- Every result carries the Preliminary label. Final classification stays with your organization and its legal counsel.
Your app records each disclosure as it shows it.
The careers site shows the notice, then posts one record per conversation to the REST API. The record lands in the system's Article 50 documentation and in the next evidence snapshot.
import { createHash } from "node:crypto";
const api = `https://api.connic.co/v1/projects/${PROJECT_ID}`;
const system = `${api}/compliance/systems/${SYSTEM_ID}`;
export async function recordDisclosure(runId: string, reply: string) {
await fetch(`${system}/transparency-applications`, {
method: "POST",
headers: {
Authorization: `Bearer ${process.env.CONNIC_API_KEY}`,
"Content-Type": "application/json",
},
body: JSON.stringify({
idempotency_key: `careers-chat-${runId}`,
run_id: runId,
disclosure_type: "direct_interaction",
status: "applied",
surface: "careers site chat",
output_hash: createHash("sha256").update(reply).digest("hex"),
provenance: {
source: "customer_attestation",
attested_by: "Careers site team",
attested_at: new Date().toISOString(),
evidence_reference: `banner-v3/${runId}`,
basis: "Banner shown before the first reply.",
},
}),
});
}Authorization: `Bearer ${process.env.CONNIC_API_KEY}`,One key, one permission
An API key with
compliance.managecan record evidence. Keep it on your server, never in the browser.idempotency_key: `careers-chat-${runId}`,Safe to retry
A repeated request with the same
idempotency_keyand content returns the existing record. Reusing the key with other content fails.output_hash: createHash("sha256").update(reply).digest("hex"),A hash instead of the reply
run_idties the record to the run, andoutput_hashstores a SHA-256 of what the candidate saw.source: "customer_attestation",Your app attests
connic records the notice your app showed. It leaves prompts and replies unchanged and adds no notices itself.