EU AI Act Compliance: Risk Tiers, Duties, and the Amended Dates

Does the EU AI Act apply to a company outside the EU?

Often, yes: the EU AI Act can cover a company with no EU presence if it sells an AI system into the EU or if its system’s output is used there.

Article 2 covers providers that place AI systems or general-purpose AI models on the EU market, wherever the provider is established. It covers deployers located in the EU. It also covers providers and deployers in third countries where the output of the system is used in the EU.

That last clause is the one US companies miss. A hiring tool run from Texas that scores candidates for a job in Ireland can fall inside the Act even though no server sits in Europe.

Am I a provider or a deployer?

As a rule of thumb, a company that sells AI is a provider and a company that buys and uses it is a deployer, and that split decides almost every duty. The legal definitions are more precise than the shorthand.

A provider develops an AI system or model, or has one developed, and places it on the market or puts it into service under its own name or trademark. A deployer uses an AI system under its authority in a professional activity.

The trap is in between. Under Article 25, a deployer can become a provider of a high-risk system if it puts its own name on the system, makes a substantial modification, or changes the intended purpose so the system becomes high-risk.

A company that rebrands a vendor’s model and sells it to its own clients should assume it has taken on provider duties.

What are the risk tiers?

The Act sorts AI into prohibited practices, high-risk systems, systems with transparency duties, general-purpose AI models, and minimal-risk systems, and the duties scale with the tier.

Tier	What it covers	Main consequence
Prohibited (Article 5)	A closed list of practices considered unacceptable	Banned. Highest fines
High-risk (Article 6, Annexes I and III)	Safety components of regulated products, and listed uses such as employment, credit scoring, education, essential services, law enforcement, migration	Full set of provider duties, plus deployer duties
Transparency (Article 50)	Systems that interact with people, generate synthetic content, recognize emotions, or create deepfakes	Disclosure and marking duties
General-purpose AI models (Articles 51 to 55)	Models that can perform a wide range of tasks, with added duties for models with systemic risk	Documentation, copyright policy, training data summary; evaluations and incident reporting for systemic-risk models
Minimal risk	Everything else	No new obligations beyond AI literacy measures

Which practices are prohibited?

Article 5 bans a closed list of practices, and the ban has applied since February 2, 2025. The list includes:

• Manipulative or deceptive techniques that materially distort behavior and cause significant harm
• Exploiting vulnerabilities linked to age, disability, or social or economic situation
• Social scoring that leads to unjustified or disproportionate treatment
• Predicting an individual’s risk of committing a crime based solely on profiling or personality traits
• Building facial recognition databases through untargeted scraping of images
• Emotion recognition in the workplace and in education, except for medical or safety reasons
• Biometric categorization that infers sensitive traits such as race, political opinions, or sexual orientation
• Real-time remote biometric identification in public spaces for law enforcement, subject to narrow exceptions

The Digital Omnibus added two prohibitions that apply from December 2, 2026: systems that generate or manipulate intimate or sexually explicit images of identifiable people without their explicit consent, and systems that generate child sexual abuse material. For providers, the ban applies where that output is the system’s intended purpose, or a reasonably foreseeable outcome the system lacks adequate safeguards to prevent; for users, it applies where they use the system to produce such material (Article 5(1a)).

The workplace emotion recognition ban is the one most likely to catch an ordinary business. Interview tools and call-center software that score mood or sentiment from a worker’s face or voice deserve a close look.

What do general-purpose AI model providers owe?

Since August 2, 2025, providers of general-purpose AI models must keep technical documentation, give information to downstream providers that build on the model, put in place a policy to comply with EU copyright law, and publish a summary of the content used for training. Providers of models released under a free and open-source license are exempt from the first two of those duties unless the model has systemic risk (Article 53(2)).

Models with systemic risk carry more. A model is presumed to have systemic risk when its training compute exceeds 10²⁵ floating-point operations. Those providers must evaluate the model, assess and mitigate systemic risks, report serious incidents, and maintain cybersecurity protection.

The Commission published a voluntary General-Purpose AI Code of Practice in July 2025. Signing it is one way to show compliance; it is not the only way.

Two dates matter for enforcement. The Commission’s page states that national authorities and the AI Office are responsible for supervising and enforcing the Act from August 2, 2026. Models placed on the market before August 2, 2025 have until August 2, 2027 to comply.

A company that only calls a model through an API is usually not the model provider. It may still be the provider of the AI system it builds on top.

What does high-risk compliance require?

High-risk compliance means a full product compliance program for providers and a shorter list of operating duties for deployers. Providers need:

• A risk management system that runs across the life of the system (Article 9)
• Data governance for training, validation, and testing data (Article 10)
• Technical documentation (Article 11) and automatic logging (Article 12)
• Instructions for use that explain capabilities and limits (Article 13)
• Human oversight measures built into the design (Article 14)
• Accuracy, robustness, and cybersecurity (Article 15)
• A quality management system (Article 17)
• Conformity assessment, CE marking, and registration in the EU database before market placement
• Post-market monitoring and serious incident reporting

For deployers, Article 26 requires use according to the instructions, human oversight by trained staff, monitoring of operation, keeping the logs under their control for at least six months, and informing workers before a high-risk system is used in the workplace.

Some deployers must also complete a fundamental rights impact assessment under Article 27. That includes public bodies, private entities providing public services, and deployers that use AI for creditworthiness evaluation or for life and health insurance risk assessment and pricing.

The Omnibus made several changes inside these duties. It allows the fundamental rights impact assessment to cross-reference a GDPR data protection impact assessment, restructures the training data rules in Article 10, and adds simplified documentation for small and mid-sized companies. Read the consolidated text, not a 2024 summary.

What exactly did the Digital Omnibus change?

The Digital Omnibus delayed most high-risk duties without delaying the rest of the Act, and made a set of narrower changes to the duties themselves. It is Regulation (EU) 2026/1744, published in the Official Journal on July 24, 2026, and it entered into force on July 27, 2026.

The Regulation gives the reason: standards and the national governance and conformity assessment setup were delayed, which made the original high-risk date unworkable (Regulation (EU) 2026/1744, recital 2). The requirements themselves stayed.

Other changes include:

• AI literacy under Article 4 now requires measures to support literacy, without guaranteeing a level for any individual
• A new Article 4a that moves the narrow exception for processing special-category personal data for bias detection out of Article 10 and extends it, with safeguards, beyond high-risk providers
• A grace period to December 2, 2026 for the Article 50(2) marking duty on generative systems placed on the market before August 2, 2026
• Simplified technical documentation and quality management for small and mid-sized companies, and lower fine caps for small mid-caps
• Expanded AI Office powers over certain systems built on general-purpose models and systems inside very large online platforms

What are the key dates?

The next deadlines are December 2, 2026, for the new prohibitions and the end of the marking grace period, and December 2, 2027, for stand-alone high-risk systems. The full sequence:

Date	What applies
August 1, 2024	AI Act enters into force
February 2, 2025	Prohibited practices and AI literacy
August 2, 2025	General-purpose AI model obligations; governance and penalty provisions (except the Commission’s fines on general-purpose AI model providers under Article 101, which apply from August 2, 2026)
July 27, 2026	Digital Omnibus on AI enters into force
August 2, 2026	General date of application, including Article 50 transparency duties; AI Office and national authorities enforce
December 2, 2026	New prohibitions on non-consensual intimate imagery and child sexual abuse material; end of Article 50(2) marking grace period for systems already on the market
August 2, 2027	Deadline for general-purpose AI models placed on the market before August 2, 2025
December 2, 2027	High-risk duties for stand-alone Annex III systems (was August 2, 2026)
August 2, 2028	High-risk duties for AI in Annex I regulated products (was August 2, 2027)
August 2, 2030	Deadline for providers and deployers of high-risk systems intended for public authorities that were already on the market before the high-risk rules applied (Article 111(2)); new systems must comply when the high-risk rules apply

How large are the penalties?

The largest EU AI Act fine is up to €35 million or 7% of worldwide annual turnover, for prohibited practices. Article 99 sets three ceilings, and for each the higher of the two figures applies.

• Prohibited practices: up to €35 million or 7% of worldwide annual turnover
• Most other obligations, including high-risk and transparency duties: up to €15 million or 3%
• Supplying incorrect or misleading information to authorities: up to €7.5 million or 1%

For small and medium-sized enterprises, the lower of the two figures applies. Article 101 sets a separate cap for general-purpose AI model providers of €15 million or 3%.

These are ceilings. The Act directs authorities to weigh the nature, gravity, and duration of the violation and the size of the company.

What should a US company do now?

Work on what already applies first. The Omnibus bought time for high-risk systems, but the prohibitions, the general-purpose model duties, and the transparency rules are in effect now.

• Map which systems are placed on the EU market or produce output used in the EU
• Decide, system by system, whether you are a provider, a deployer, or both
• Confirm that no system uses a prohibited practice, especially emotion recognition at work
• Meet Article 50 duties now: tell people when they are talking to AI, and label deepfakes
• For high-risk systems, use the delay to build documentation and testing rather than to pause
• If you build on a general-purpose model, collect the provider’s downstream documentation
• Track the consolidated text and Commission guidelines, which are still being issued

For more on how the EU rules sit next to US law, see the AI compliance guide and governance frameworks.

This guide is general information as of October 2026, not legal advice. Confirm obligations against the consolidated text of the Act and with EU counsel.