EU AI Act Is Live August 2, 2026. Here's What Developers Need to Build.

Article 14 of the EU AI Act mandates “human oversight” for high-risk AI systems. The 144-page regulation boils down to three requirements:

1. A human must understand the AI's output — your system needs to explain what it did and why
2. A human must be able to override or stop the AI — there must be a mechanism to intervene before irreversible actions
3. The oversight must be proportionate to risk — a chatbot suggesting restaurants needs less oversight than an AI approving insurance claims

The regulation doesn't tell you how to implement oversight. It tells you thatyou must. The implementation is your problem. Fines for non-compliance: up to €35 million or 7% of global annual turnover, whichever is higher.

If you answer “yes” to any of these, Article 14 applies to you:

Even if your system isn't “high-risk” under Annex III, adding oversight is good practice. AI agents that send emails, process refunds, or modify user data should have guardrails regardless of legal requirements.

Article 14 compliance requires three technical components:

1. A checkpoint system

Before your AI takes an irreversible action, it pauses and waits for human approval. The human sees the full context — what the AI wants to do, why, and what data it used. They approve, deny, or modify the action. If nobody reviews within a timeout period, the action is automatically denied (never auto-approved).

2. An audit trail

Every AI decision must be logged: what was decided, when, by whom, and based on what data. Records must be retained for the lifetime of the AI system plus 5 years. When the regulator asks “Why was this loan denied?”, you must answer with specifics — not “the model said so.”

3. A notification system

Humans can't oversee what they don't see. Your system must actively notify reviewers when their input is needed — via Slack, email, or in-app. A dashboard that nobody checks isn't oversight.

ApprovalGate implements all three components — checkpoints, audit trail, and Slack notifications — with minimal code changes.

Python

from luxkern import ApprovalGate

gate = ApprovalGate(api_key="lxk_live_xxx")

def process_loan(application):
    decision = ai_model.evaluate(application)
    result = gate.checkpoint(
        action="loan_decision",
        context={
            "applicant_id": application.id,
            "amount": application.amount,
            "ai_decision": decision.outcome,
            "confidence": decision.confidence,
        },
        timeout_minutes=30,
    )
    if result.approved:
        execute_decision(decision)
    else:
        flag_for_manual_review(application)

Node.js

import { ApprovalGate } from "@luxkern/sdk";

const gate = new ApprovalGate({ apiKey: "lxk_live_xxx" });

async function processRefund(order) {
  const result = await gate.request({
    action: "process_refund",
    agent: "refund-bot",
    context: {
      order_id: order.id,
      amount: order.total,
      customer: order.email,
    },
    timeoutMinutes: 15,
  });

  if (result.approved) {
    await stripe.refunds.create({
      payment_intent: order.paymentId,
    });
  }
}

When a checkpoint is created, the reviewer gets a Slack notification with the full context. They click Approve or Deny directly from Slack. The entire review takes 8-15 seconds. Everything is logged automatically — every decision, every timestamp, every reviewer.