EU AI Act: Know Which AI You Run, and What It Obliges You To Do
Regulation (EU) 2024/1689 applies in phases, with prohibited practices and AI literacy already in force since February 2025 and general purpose AI obligations since August 2025. Most organisations still cannot list the AI systems they use, let alone classify them.
EU AI Act Compliance. Regulation (EU) 2024/1689 applies in phases, with prohibited practices and AI literacy already in force since February 2025 and general purpose AI obligations since August 2025. Most organisations still cannot list the AI systems they use, let alone classify them.
Does This Sound Familiar?
No inventory of AI systems
AI has arrived through SaaS features, copilots, recruitment tools and models built by your own teams. Nobody holds a single list, so no obligation can be assigned.
Unclear role and risk class
Provider, deployer, importer or distributor changes everything, and so does whether a system is prohibited, high-risk, limited-risk or minimal-risk. Getting this wrong means either paralysis or exposure.
Governance and evidence are missing
High-risk systems require a risk management system, data governance, technical documentation, logging, human oversight and post-market monitoring. Very few organisations can evidence any of these today.
Our EU AI Act Compliance Methodology
1.
Inventory every AI system in use
We identify AI systems across business units and suppliers, including embedded AI features and internal models, and record purpose, data and decision impact for each.
2.
Determine role and classification
We establish whether you act as provider or deployer for each system and classify it against the prohibited, high-risk, transparency and minimal-risk categories in the Regulation.
3.
Design proportionate AI governance
An AI policy, an approval gate for new systems, roles and responsibilities, an AI literacy plan for staff, and the controls required for the systems that actually carry risk.
4.
Build the compliance evidence
Technical documentation, data governance records, logging, human oversight arrangements, transparency notices and post-market monitoring for high-risk systems, aligned with your ISO 27001 or ISO 42001 work.
Tangible Deliverables
AI system register
A complete register of AI systems with owner, purpose, role, risk class and applicable obligations.
Classification and gap report
A documented classification per system with the gaps against the applicable articles and the phased deadlines.
AI governance framework
AI policy, intake and approval process, RACI, supplier requirements and an AI literacy programme meeting Article 4.
High-risk compliance pack
Risk management system, technical documentation template, logging and human oversight design, and post-market monitoring plan.
Real Results, Not Just Reports
2026
high-risk deadline
Most high-risk obligations apply from 2 August 2026, with certain embedded product cases running to 2027.
4
risk categories
Prohibited, high-risk, limited-risk with transparency duties, and minimal-risk. Your obligations follow the class, not the hype.
6-10
weeks to readiness
A typical inventory, classification and governance engagement for a mid-sized organisation.
In Practice
Challenge
A Belgian services group discovered fourteen AI features across HR, marketing and customer support tools, with one recruitment screening use case that was potentially high-risk and no governance at all.
Approach
DigiSôter inventoried and classified every system, confirmed the deployer role, retired two uses that could not be justified, and built an AI policy, an intake gate and an AI literacy programme.
Outcome
The group now approves new AI uses in days rather than months, holds documented classifications for each system, and has human oversight and logging in place for the single high-risk use case.
Frequently Asked Questions
Yes. If you use an AI system under your own authority you are a deployer, and deployers carry obligations such as AI literacy, human oversight for high-risk systems, transparency towards affected people and use in line with the provider instructions. Building or substantially modifying a system makes you a provider, with far heavier duties.
Prohibited practices and the AI literacy obligation have applied since 2 February 2025, obligations for general purpose AI models since 2 August 2025, most high-risk system obligations from 2 August 2026, and certain AI embedded in regulated products from 2 August 2027.
We inventory every AI system in use, record its purpose and the decisions it influences, establish your role for each, and then classify it against the prohibited practices, the high-risk annexes, the limited-risk transparency cases and minimal risk. Each classification is documented so it can be defended later.
A risk management system across the lifecycle, data and data governance requirements, technical documentation, automatic logging, transparency and instructions for use, human oversight, and appropriate accuracy, robustness and cybersecurity, plus post-market monitoring and incident reporting.
Yes. Risk management, supplier control, change management and evidence discipline are reused directly, and the AI governance framework can sit alongside an ISO 42001 AI management system if you decide to certify later.
Your biggest security perimeter is no longer your firewall. It is your suppliers. In a world of cloud, outsourcing, and interconnected operations, trust is essential, but trust without verification is not governance. It is hope.
Before you read any further, test your governance in 10 seconds: Who can access your plant remotely, under which conditions, and how fast can you cut them off? If the answer is not immediate, you already have an exposure problem.