EU AI Act: what it means for coaching tools

From August 2026, AI systems used in employment contexts with identifiable individuals are high-risk under EU law. FirstLineCoach is designed to comply from day one.

What the regulation says

EU AI Act — Annex III, Point 4(b)

AI systems intended to be used to make or materially influence decisions affecting the terms of work-related relationships, including task allocation based on individual behaviour, personal traits, or characteristics — are classified as high-risk.

This applies from August 2026. Systems already in use will have a transition period, but new deployments must comply from the start.

What this means in practice

Most coaching and QA tools that use AI to recommend coaching approaches for specific, identifiable people will need to comply with the full set of high-risk requirements: risk management systems, data governance, technical documentation, human oversight, accuracy and robustness standards, and a conformity assessment.

This is not a minor compliance exercise. High-risk classification carries significant obligations for providers and deployers.

FirstLineCoach's architectural boundary

We don't avoid AI — we use it carefully, with a clear boundary between what AI touches and what it doesn't.

AI is used here

Voluntary, post-session, no individual targeting
  • Vignette capture (/reflect) — After a coaching session, coaches voluntarily capture their reflections. AI structures these into coaching vignettes. The coach reviews and approves before anything is stored.
  • Editorial triage (/triage) — AI reviews draft coaching content for quality and consistency. This is editorial, not employment-related.

AI is never used here

Deterministic, no AI at runtime
  • Plan library and search — The coaching plans and the scenario search are entirely deterministic. Search phrases are generated at build time; matching at runtime uses no AI whatsoever.
  • Session preparation — Coaches prepare for sessions using the plan library and loop diagram. No AI recommends an approach for a specific individual.
  • The loop diagram — A diagnostic tool for use during conversations. No AI involvement.

How we enforce the boundary

Automated drift guard

An automated test suite (AiDriftGuardTests) runs in CI and enforces the boundary in the codebase. If someone introduces an AI call in a context that would cross the Annex III 4(b) line, the build fails.

DPIA support pack

A full Data Protection Impact Assessment support pack and compliance documentation has been prepared, covering data flows, processing purposes, retention policies, and the specific architectural decisions that keep us outside the high-risk boundary.

Architectural documentation

The boundary is documented in the project's architecture documents. Any change to where AI is used requires explicit review against the EU AI Act classification criteria.

The key distinction

The question is not whether you use AI — it's whether AI is used to make or influence decisions about identifiable individuals in an employment context.

FirstLineCoach uses AI to help coaches reflect on their own practice (voluntary, post-session) and to review editorial content (not employment-related). It never uses AI to determine what coaching approach to take with a specific person. That's the line.

Want to understand how this affects your organisation?

If you're using AI-powered coaching or QA tools, the August 2026 deadline matters. We're happy to discuss how the regulation applies to your specific setup — no sales pitch, just clarity.

Get in Touch