---
title: Does the EU AI Act apply to my US AI startup?
description: When the EU AI Act reaches a US company, the risk tiers, what applies now (prohibited practices, general-purpose model rules, and transparency duties), the delayed high-risk deadlines after the 2026 Digital Omnibus, and what to do first.
category: AI features
service: ai-feature-review
updated: 2026-11-17
---

## Short answer

It can, even if you have no office in Europe. The EU AI Act applies to companies that offer AI systems in the EU, and to AI systems whose output is used there. For most US startups building on existing models, what applies today is a short list: a set of prohibited uses, and transparency duties such as telling people they're talking to an AI and marking AI-generated content. The heavier requirements for high-risk uses, such as hiring, credit, and education, were delayed in 2026 and now start in December 2027. Knowing which tier your product falls into tells you how much of the law you actually need to plan for.

## Does it reach you?

- **You offer AI in the EU.** If you place an AI system on the EU market or put it into service there, the law applies wherever your company is based.
- **Your AI's output is used in the EU.** It also applies to providers and deployers outside the EU when the system's output is used in the EU.
- **Your role.** A **provider** develops an AI system or has one developed and offers it under its own name. A **deployer** uses an AI system in its business. A startup that builds an AI product on a model from OpenAI, Anthropic, or Google is usually the provider of that product, while the model company is the provider of the underlying general-purpose model.

## What applies, by tier

### Prohibited practices (since February 2025)

A short list of uses is banned outright, including AI that manipulates people or exploits vulnerabilities such as age or disability in ways that cause significant harm, social scoring, building facial recognition databases by untargeted scraping, emotion recognition in workplaces and schools except for medical or safety reasons, and predicting crime based solely on profiling. Most products are nowhere near these, but check features that infer emotions or target vulnerable users.

### General-purpose AI models (since August 2025)

Companies that provide general-purpose models, such as large language models, have documentation, copyright policy, and training data summary duties. If you build on someone else's model, these duties generally sit with the model provider. They can shift to you if you substantially modify a model and offer it in the EU. See [Can I use an open-weight model like Llama or Gemma in my product?](/guides/using-open-weight-models-commercially)

### Transparency duties (since August 2026)

- **Chatbots and agents** must make clear that people are interacting with an AI, unless it's obvious. See [Do I have to tell users they're talking to an AI?](/guides/telling-users-they-are-talking-to-ai)
- **AI-generated content.** Providers of systems that generate images, audio, video, or text must mark outputs in a machine-readable way so they can be detected as AI-generated. Systems already on the market before August 2, 2026 have until December 2, 2026 to add this marking.
- **Deepfakes** must be disclosed by the deployer, with lighter rules for clearly artistic or satirical work.
- **Emotion recognition and biometric categorization** require notice to the people exposed to them.

### High-risk systems (delayed to December 2027)

High-risk uses include AI used in hiring and managing workers, education and exam scoring, access to essential services such as credit scoring and pricing of life and health insurance, biometrics, critical infrastructure, and certain government uses. Providers of these systems must have risk management, data governance, technical documentation, logging, human oversight, accuracy and security testing, a conformity assessment, and EU registration. Non-EU providers must also appoint an authorized representative in the EU.

The Digital Omnibus on AI, Regulation (EU) 2026/1744, in force since July 27, 2026, moved these obligations from August 2, 2026 to December 2, 2027, and to August 2, 2028 for AI built into products already covered by EU safety laws, such as medical devices and machinery. A system in a high-risk area can fall outside the tier if it doesn't pose a significant risk, for example because it performs only a narrow procedural task, but you have to document that assessment. See [Can my AI product screen job candidates or evaluate workers?](/guides/ai-hiring-and-employment-tools)

## Penalties

Fines can reach €35 million or 7% of worldwide annual turnover for prohibited practices, and €15 million or 3% for most other violations. For small and medium-sized companies, the lower of the two amounts applies.

## What to do first

- **Map where your product is used.** Know whether you have EU users or customers, or whether customers use your outputs in the EU.
- **Classify each AI feature** as prohibited, high-risk, transparency-only, or minimal risk, and write down why.
- **Meet the transparency duties now:** AI disclosures in chat and voice products, and machine-readable marking of generated media.
- **If anything is high-risk,** use the time before December 2027 to build documentation, logging, testing, and human oversight into the product, since enterprise customers in the EU are already asking about them.
- **Remember the GDPR.** The AI Act sits alongside EU privacy law, which already applies to personal data you process from people in the EU.

## Checklist an agent can run against the codebase

- [ ] Find whether the product serves EU users or customers: locales, billing countries, data regions, and geo settings.
- [ ] List every AI feature and what it decides, generates, or infers.
- [ ] Flag features in high-risk areas: hiring and worker management, education, credit and insurance, and biometrics.
- [ ] Find any emotion recognition, biometric categorization, or features aimed at children or other vulnerable users.
- [ ] Check that chat and voice interfaces disclose that the user is talking to an AI.
- [ ] Check whether generated images, audio, video, and text carry machine-readable marking or provenance metadata.
- [ ] Check for logging of AI inputs, outputs, and model versions that would support documentation and oversight.
- [ ] Check whether any model is fine-tuned or substantially modified before being offered to customers.

## Sources

- [EU AI Act, Regulation (EU) 2024/1689](https://eur-lex.europa.eu/eli/reg/2024/1689/oj)
- [Digital Omnibus on AI, Regulation (EU) 2026/1744](https://eur-lex.europa.eu/eli/reg/2026/1744/oj)
- [European Commission: AI Act Service Desk](https://ai-act-service-desk.ec.europa.eu/en)

## If this applies to you

An **AI feature review** classifies your AI features under the EU AI Act, covers the transparency duties that apply now, and identifies anything in a high-risk area with a plan for the December 2027 deadline, alongside the US laws that apply. For one feature, such as adding emotion analysis or AI-generated media, a **pre-build feature check** covers it before you build.
