Martin KogutBy Martin KogutFebruary 11, 20265 min read

The EU AI Act. A Practical Guide for German Businesses.

The EU AI Act. A Practical Guide for German Businesses.

Updated November 2026 · 15 min read

What changed in 2026. The Digital Omnibus on AI took effect on 27 July 2026. It revised Article 4 on AI literacy and deferred key high-risk requirements to December 2027 or August 2028, depending on the system. Article 50 transparency duties generally apply from August 2026. The timeline covers the transition for certain synthetic-content tools and the new December 2026 prohibitions.

Why oversight matters

Reporting in 2018 described how Amazon abandoned an experimental recruitment tool. Trained on historical applications, the model learned to penalise some terms associated with women, including references to women’s clubs.

HireVue removed facial analysis from its video assessment products after criticism of the technique and its risks for candidates. In a separate US case, Derek Mobley alleged that Workday’s screening software discriminated in job applications. A May 2025 ruling allowed a preliminary collective action over alleged age discrimination to proceed.

These cases illustrate the stakes of oversight. The EU AI Act sets obligations based on the role, system and risk. Penalties vary by violation and business size.

Who should pay attention?

In a September 2026 Bitkom survey, 57% of German businesses with at least 20 employees said they used AI, up from 36% a year earlier. If your team drafts emails, screens applications or publishes synthetic media, identify which systems it uses and what obligations apply.

AI can influence decisions with lasting effects. In the Dutch childcare benefits scandal, discriminatory fraud risk scoring was one part of a wider process that wrongly accused and harmed families. The case illustrates why teams need to know how decisions are made and how to challenge or correct them.

The EU AI Act sets rules for uses that can affect people’s health, safety or fundamental rights. Start by identifying where your teams use AI and who the outputs affect.

The timeline: what applies when?

The rules apply in stages. Check the dates against your role and how your system is used.

1 August 2024. The EU AI Act entered into force.

2 February 2025. The first Article 5 prohibitions and Article 4 on AI literacy began to apply. Revised Article 4 took effect on 27 July 2026 and requires providers and deployers to take context-appropriate measures that support AI literacy.

2 August 2025. Penalty provisions and obligations for providers of general-purpose AI models began to apply, subject to the relevant transition provisions. Germany’s national implementation law took effect on 29 July 2026.

2 August 2026. Article 50 transparency duties generally began to apply. Providers of synthetic-content systems placed on the market before this date have until 2 December 2026 to meet Article 50(2)’s marking requirements.

2 December 2026. New Article 5 prohibitions begin to apply to specified uses and provision of AI systems involving non-consensual intimate material or child sexual abuse material. This is also the end of the Article 50(2) transition described above.

2 December 2027. Key requirements begin to apply to systems classified as high-risk under Article 6(2) and Annex III, including certain uses in recruitment, credit and education.

2 August 2028. Key high-risk requirements begin to apply to systems classified under Article 6(1) and Annex I, including qualifying AI safety components in regulated products.

Which practices are prohibited?

Article 5 prohibits specific practices. Check its conditions and exceptions when assessing a system.

Social scoring

Article 5 prohibits defined forms of social scoring based on behaviour or personal characteristics when it leads to adverse treatment in an unrelated context, or treatment that is unjustified or disproportionate.

Emotion recognition at work or in education

Inferring emotions from biometric data at work or in education is prohibited, with an exception for medical or safety reasons.

Biometric categorisation by sensitive characteristics

The Act prohibits specified forms of biometric categorisation that infer sensitive characteristics such as political opinions, religion or sexual orientation. Article 5 sets out the relevant exclusions.

Manipulation and exploitation of vulnerabilities

The prohibition covers specified subliminal, manipulative or deceptive techniques and the exploitation of vulnerabilities such as age, disability or economic circumstances. The law sets conditions concerning distorted behaviour and significant harm.

Untargeted scraping of facial images

Article 5 prohibits systems that create or expand facial recognition databases by untargeted scraping of facial images from the internet or CCTV.

For a breach of Article 5, Article 99 sets a maximum of €35 million or 7% of the previous year’s worldwide annual turnover for an undertaking, whichever is higher. For small and medium-sized enterprises, including start-ups, the lower cap applies.

The AI literacy obligation

Article 4 requires providers and deployers to take measures that support the AI literacy of staff and others who operate or use AI systems on their behalf.

Match your measures to people’s knowledge, roles, systems and the people affected by their use. Someone drafting an email and someone reviewing a recruitment decision need different guidance.

The revised rules give teams flexibility in how they develop and document AI literacy. Keep an internal record of training and other measures, and review them when the systems or work change.

See our guide to Article 4 for practical measures and the Commission’s current guidance.

When AI use is high-risk

The Act classifies certain intended uses as high-risk under Article 6 and Annexes I and III. Check the specific use and any applicable exceptions.

Recruitment and HR

Systems intended to filter applications, evaluate candidates or support certain employment decisions are listed in Annex III. The classification depends on the intended use and Article 6 criteria, including its rules on profiling.

Credit scoring

Annex III lists AI intended to evaluate a person’s creditworthiness or establish a credit score, with an exception for detecting financial fraud.

Education and vocational training

Examples in Annex III include evaluating learning outcomes, deciding admission and determining access to education or vocational training.

Critical infrastructure

Annex III includes certain AI systems used as safety components in managing critical digital infrastructure, road traffic and supplies of water, gas, heating or electricity.

Law enforcement and justice

Annex III identifies specified uses in law enforcement, administration of justice and democratic processes. Review the exact task and the authorities involved.

For applicable high-risk systems, providers may need risk management, data governance, technical documentation, conformity assessment and registration. Deployers have their own duties, including steps for human oversight and monitoring. Applicability depends on the classification, role and transition provisions.

Transparency duties under Article 50

Article 50 applies to specified interactions and types of synthetic or manipulated content. It assigns different duties to providers and deployers.

Tell users when they are interacting with AI

Providers of systems designed to interact directly with people must ensure users are told they are interacting with AI, unless it is obvious in the circumstances or a legal exception applies.

Mark synthetic media and disclose deepfakes

Providers of systems that generate synthetic audio, images, video or text generally have a duty to make outputs machine-readable as generated or manipulated. A separate disclosure duty applies to deployers publishing deepfakes, subject to the article’s exceptions and provisions for creative works.

Synthetic text in certain contexts

Deployers generally disclose AI-generated or altered text published to inform the public on matters of public interest. An exception applies where there is human review or editorial control and someone takes editorial responsibility for publication.

Check what each tool produces and which organisation must mark or disclose it. For an existing synthetic-content tool, confirm whether Article 50(2)’s transition applies.

Germany’s AI Service Desk

The Federal Network Agency’s AI Service Desk opened in July 2025. Under Germany’s implementation law, effective 29 July 2026, the agency also has a central coordination and market surveillance role. Existing sector regulators retain responsibilities in areas such as finance.

The Service Desk offers four key resources:

The Compliance Compass: An online tool that helps you determine whether and how the AI Act applies to your AI systems. You answer a few questions, and the system shows you which obligations you face. It's no substitute for legal advice, but it's a good starting point.

Information resources: FAQs, examples, and explanations covering all aspects of the AI Act. Particularly valuable: concrete scenarios that show where the line between permitted and prohibited lies.

Direct contact: You can submit questions directly to the authority. This is especially useful for companies operating in a grey area who need clarity.

Training guidance: Links to free training resources that can support your Article 4 measures.

Visit the Federal Network Agency’s AI Service Desk for guidance and contacts in Germany.

What your business can do now

Start with the systems your team already uses.

1. Conduct an inventory

Create an inventory of approved AI systems and tools staff use on their own. Record each system’s purpose, users and outputs so you can check which obligations apply.

2. Clarify your role

Check your role for each system. You may be a provider if you develop or commission a system and market it or put it into service under your name. If you use a system under your authority, you may be a deployer. Branding, substantial modification or a change of intended purpose can change your obligations.

3. Check for prohibited practices

Review any use involving emotion recognition, manipulation, sensitive biometric categorisation or other Article 5 practices. Take uncertain classifications to the people responsible for legal review.

4. Start AI training

Plan AI literacy measures around the tools and roles in your inventory. Teach people to check results and escalate problems, then document what you put in place.

5. Prepare for transparency obligations

Review user notices for direct AI interactions, provider marking for synthetic outputs and deployer disclosures for the content you publish. Check which conditions and dates apply.

6. Plan ahead for high-risk AI

If a use may be high-risk, assess it against Article 6 and the applicable annex. Identify who is responsible, which obligations apply and how long documentation and operational controls will take.

The opportunity behind the regulation

An AI inventory can make it easier to see who owns a system and where its outputs affect people.

The Amazon, HireVue and Workday examples illustrate different ways questions can arise about an automated decision. Document who reviews the system, how outcomes are assessed and where people can raise concerns.

Use those records to explain to staff, customers and partners how AI is used in your organisation and who is responsible.

Review the process when you change models, add features or introduce new uses.

Keep the evidence with the decision.

Start with the rules in force

AI literacy, the initial prohibited practices and applicable transparency duties are already in effect.

Start with your inventory and confirm how your systems fit the rules. Use the Federal Network Agency’s guidance and involve the responsible specialists when you need a legal assessment.

Use the staggered deadlines to assign owners and schedule the work that applies to your systems.


Hypescale helps teams map AI use, assess deployment options and put agreed governance controls into practice. Your legal and regulatory specialists determine which duties apply.

Discuss your AI strategy