The EU AI Act Deadline Nobody On Your Team Has Diarised

The AI Act's obligations arrive in stages, and the stage most organisations overlook is the one that applies to them as deployers rather than developers.

If your organisation uses an AI system for recruitment, credit decisions, or performance evaluation, you are in scope even though you did not build the model.

What deployers actually owe

• Human oversight. A named person must be able to interpret and override the system's output.
• Staff competence. Anyone operating the system needs demonstrable AI literacy. This is the clause that turns training from nice-to-have into evidence you may be asked to produce.
• Logging. Keep the system's automatically generated logs for the required retention period.
• Transparency to affected people. Tell candidates and employees when an AI system is materially involved in a decision about them.

Australia is not far behind

The Australian framework remains principles-based rather than binding, but the proposed mandatory guardrails for high-risk settings mirror the EU's structure closely. Organisations already building the EU evidence trail will not have to start again.

The practical takeaway: whatever you decide about the policy, start keeping a record of who has been trained and on what. That record is the cheapest thing to build now and the most expensive to reconstruct later.