Your trading algorithms are not high-risk under the AI Act. They are, however, already governed by MiFID II and RTS 6 — which is why nobody should be rebuilding that governance from scratch.
Check your exposure — freeAnnex III does not mention algorithmic trading, best execution, portfolio optimisation or investment advice. Not once. An investment firm can run its entire AI estate and touch the AI Act's high-risk chapter only through the side door — as an employer, or if it lends to retail clients. What governs your algos is MiFID II Article 17 and RTS 6, and ESMA has already told you how it reads AI against them.
Read this the way the Act reads it: Annex III is a list of named use cases, not a judgement about how important a system is to your business. The most consequential model you run may well be the one the Act says nothing about.
| System | Under the AI Act | Why |
|---|---|---|
| Algorithmic and high-frequency trading | Not high-risk Not high-risk — but see MiFID II | Absent from Annex III. Governed instead by MiFID II Art. 17 and RTS 6 (Delegated Regulation (EU) 2017/589): governance, pre-deployment testing, kill functionality, annual self-assessment. ESMA published a supervisory briefing on algorithmic trading in February 2026. |
| Robo-advice and automated suitability assessment | Not high-risk Not high-risk | Not named in Annex III. Governed by the MiFID II suitability regime — and by ESMA's Public Statement on AI and investment services of 30 May 2024, which expects organisational rigour, transparency about AI's role, and the duty to act in the client's best interest. |
| Best execution and smart order routing | Not high-risk Not high-risk | Not in Annex III. This is MiFID II best-execution territory. |
| Market abuse surveillance | Not high-risk Not high-risk | Not in Annex III. MAR obligations, unchanged. |
| Margin lending or credit to retail clients | Applies High-risk — Annex III 5(b) | If you assess the creditworthiness of a natural person, you are in 5(b) — whatever the wrapper. This is the one route by which a pure investment firm lands in the high-risk chapter. |
| Generated research, client reporting, marketing copy | Applies Transparency — Art. 50 | Synthetic content must be machine-readably marked. The grace period ends 2 December 2026. Also a MiFID II fair-clear-and-not-misleading question. |
| CV screening and employee evaluation | Applies High-risk — Annex III point 4 | Catches you as an employer. The most common way an investment firm ends up deploying a high-risk AI system without knowing it. |
This is the sector where the AI Act is most often over-scoped. The instinct is understandable — trading is the most consequential AI in the building — but the Act simply does not reach it. Annex III is a list of use cases, and trading is not on the list.
What you should be doing instead is mapping, not rebuilding. Your RTS 6 algo governance already gives you pre-deployment testing, change control, kill switches and an annual self-assessment. Where the AI Act does apply — a retail credit line, an HR screening tool, generated client comms — you can hang those obligations off the control framework you already run, rather than standing up a parallel one.
We describe the delta. The Act itself — all 113 articles, in 24 EU languages — lives on our sister site.
Start free: run the AI Act exposure check to classify a specific system, or read the cross-obligation table showing what the AI Act adds to a DORA programme. Already working on DORA? See DORA for Investment Firms. Analytical guidance for compliance teams, not legal advice.
Take our free 5-minute assessment and get an instant DORA compliance score with personalised recommendations.