The EU AI Act for Engineers: Dates and Duties
On this page
Does the EU AI Act apply to you?
The EU AI Act regulates AI systems placed on the EU market or used in the EU, wherever the company is based. If European users touch your AI feature, assume it's relevant. It sorts obligations by risk, so the first engineering question is which bucket your system falls into.
| Category | Examples | Obligations |
|---|---|---|
| Prohibited | Social scoring, manipulative techniques, most real-time remote biometric ID in public, explicit deepfakes of real people | Banned |
| High-risk | AI in hiring, credit scoring, education access, critical infrastructure, regulated products like medical devices | Risk management, data governance, logging, human oversight, accuracy and security, documentation |
| Transparency duties | Chatbots, AI-generated or manipulated content, emotion recognition | Tell people, mark content |
| Minimal risk | Spam filters, most internal tools | No specific duties beyond AI literacy |
| General-purpose AI models | Foundation models | Duties for the model provider, not usually for you |
SupportBot, a customer-support chatbot, sits in the transparency category. A similar bot that decides who gets credit, or screens job applicants, would be high-risk.
EU AI Act timeline after the Digital Omnibus
The Act entered into force on 1 August 2024, with obligations phased in. In 2026 the EU adopted the Digital Omnibus on AI, which entered into force on 27 July 2026 and moved several dates:
| Date | What applies |
|---|---|
| 2 February 2025 | Prohibited practices. AI literacy, softened by the Omnibus to supporting staff literacy. |
| 2 August 2025 | Obligations for general-purpose AI model providers |
| 2 August 2026 | Transparency duties under Article 50 |
| 2 December 2026 | End of the grace period for machine-readable marking of AI-generated content; ban on explicit deepfakes added by the Omnibus |
| 2 December 2027 | High-risk obligations for stand-alone systems (Annex III, such as hiring and credit) |
| 2 August 2028 | High-risk obligations for AI embedded in regulated products (Annex I) |
The penalties are large: up to 35 million euros or 7 percent of worldwide turnover for prohibited practices, and up to 15 million euros or 3 percent for most other breaches. This is a summary for engineers, not legal advice, so confirm your classification with counsel.
Transparency obligations: what changes in your product
For SupportBot, Article 50 is mostly product and UI work:
- Tell people they're talking to an AI, unless it's obvious from context. A line in the chat header like "You're chatting with Acme's AI assistant" is the usual answer.
- Mark AI-generated content such as images, audio and video in a machine-readable way, for example with C2PA content credentials or watermarks, if your system generates them.
- Disclose deepfakes and AI-generated text published to inform the public, with some exceptions.
These are cheap if designed in early and annoying to retrofit.
High-risk AI systems: what engineers would build
If a system is high-risk, much of what this course teaches becomes mandatory and needs documenting:
- Risk management across the lifecycle: your threat model and risk register (Days 2 and 4).
- Data governance: provenance and quality of training and evaluation data (Days 16 and 17).
- Logging that lets incidents be traced: the tracing from Day 27.
- Human oversight: approvals and the ability to stop the system: Days 11 and 27.
- Accuracy, robustness and cybersecurity: evals and red-team results (Days 21 to 25).
- Technical documentation describing the system, its limits and how it was tested.
A practical habit: keep a system-card.md per AI feature with its purpose, risk category, model and version, data sources, known limits, human oversight, and links to the latest eval and red-team reports. It's useful for the Act, for ISO/IEC 42001 (Day 28), and for the next engineer who joins the team.
Check yourself
Article 50 transparency obligations: users must be told they're interacting with an AI system unless it's obvious from context. These obligations have applied since 2 August 2026, so a September launch must include clear disclosure.
Hiring is an Annex III use, so obligations apply from 2 December 2027 after the Omnibus. They should start risk management, data governance, logging, human oversight and documentation now, since these take time to build.
Classification can change as features grow. The card records purpose, data, limits and testing in one place, which answers regulators, customers and auditors, and keeps the team honest about what the system does.
Requirements like logging, human oversight and documentation are much cheaper to design in than to retrofit, especially for systems that may be classed as high-risk.