The EU AI Act may apply to your business's AI agents even if you're not based in the EU: what matters is whether your AI systems are offered in the EU market or their outputs are used in the EU, and what those systems are used for. The law sorts AI uses into risk tiers. Most business agents, such as internal assistants or support bots, fall into lower tiers with light duties like telling people they're talking to AI. Agents used in sensitive areas such as hiring, credit decisions or access to essential services can fall into the high-risk tier, which carries substantial obligations. This post is a general explainer, not legal advice; the rules are detailed, are applied in phases, and your situation needs checking with qualified counsel.
Below: how the risk tiers work, how to tell whether you're a "provider" or a "deployer," which agent use cases tend to land where, what non-EU companies should check, and the documentation habits worth building now regardless.
What is the EU AI Act, in plain terms?
The EU AI Act is a European Union regulation that governs how AI systems are built, sold and used, based on the risk they pose to people's health, safety and fundamental rights. Rather than regulating the technology itself, it regulates uses. The same underlying language model can power a harmless FAQ bot and a high-risk recruitment screener; the obligations attach to what the system is for.
A few features matter for businesses using AI agents:
- It's risk-based. Higher-risk uses carry heavier obligations; most everyday uses carry few or none beyond general ones.
- It assigns duties by role. Companies that develop and place AI systems on the market ("providers") carry most obligations. Companies that use AI systems in their operations ("deployers") carry a smaller but real set.
- It reaches beyond the EU. It can apply to providers and deployers outside the EU when their systems are placed on the EU market or their outputs are used in the EU.
- It applies in phases. Different obligations apply from different dates, and the timetable has been subject to discussion and proposed adjustments. Check the current timeline rather than relying on any date you've read in passing, including anywhere on this site.
It also sits alongside other laws. If your agent processes personal data of people in the EU, data protection law such as the GDPR still applies in full.
What are the EU AI Act risk tiers?
At a high level, the Act groups AI uses into four categories:
| Tier | What it covers (simplified) | What it means for you |
|---|---|---|
| Prohibited | A short list of practices considered unacceptable, such as social scoring, certain manipulative or exploitative techniques, and some uses of biometrics and emotion recognition in sensitive settings like workplaces | Don't build or use these. Check the list if any agent profiles people or infers emotions |
| High-risk | AI used in listed sensitive areas, including employment and worker management, education, access to essential private and public services such as creditworthiness, and certain safety-critical and public-sector contexts | Significant obligations: risk management, data governance, documentation, logging, human oversight, accuracy and robustness, and more |
| Transparency obligations | AI that interacts with people, generates synthetic content, or performs emotion recognition or biometric categorization | Tell people they're dealing with AI (unless obvious), and label certain AI-generated or manipulated content |
| Minimal risk | Most other business uses, such as spam filters, internal productivity tools, and many back-office automations | No specific new obligations under the Act, though general duties like staff AI literacy still apply |
General-purpose AI models, the large models many agents are built on, also carry obligations, but those fall mainly on the companies that develop the models rather than on businesses building agents on top of them.
Are you a provider or a deployer of your AI agent?
This question shapes almost everything else.
- Provider: you develop an AI system (or have it developed) and place it on the market or put it into service under your own name. A software company selling an AI recruiting tool is a provider.
- Deployer: you use an AI system under your authority in a professional context. A company using that recruiting tool to screen its own applicants is a deployer.
Businesses commonly assume they're only deployers. But if you commission a custom AI agent and put it into service under your name, or substantially modify a system, or put your brand on a system and offer it to others, you may take on provider responsibilities. This is one of the first things to clarify with counsel, especially for custom-built agents.
Which AI agent use cases are likely to be high-risk?
The tier depends on the use, so it helps to map your actual agents. A rough, non-authoritative guide:
| AI agent use case | Likely tier (depends on details) | Worth checking |
|---|---|---|
| Website or WhatsApp support agent answering product questions | Transparency obligations | Clear disclosure that customers are talking to AI |
| AI receptionist booking appointments | Transparency obligations | Disclosure, human handoff, data protection |
| Internal assistant summarizing documents or drafting emails | Minimal risk | AI literacy, data protection, internal policy |
| AI-generated marketing images or video | Transparency obligations | Labeling of synthetic or manipulated content where required |
| Agent that screens, ranks or filters job applicants | Likely high-risk | Full high-risk obligations, worker information, bias testing |
| Agent that monitors employee performance or allocates tasks based on behavior | Likely high-risk | High-risk obligations; possible prohibited practices if emotion recognition is involved |
| Agent that assesses creditworthiness or sets credit terms for individuals | Likely high-risk | High-risk obligations, possible impact assessment duties |
| Agent that decides eligibility for insurance or essential services | Possibly high-risk | Depends on the specific product and decision |
| Invoice processing or reconciliation agent | Minimal risk | General governance and data protection |
Two practical takeaways. First, most SMB agents sit in the lower tiers, and the main new duty is transparency. Second, the moment an agent influences decisions about people's jobs, money or access to services, the picture changes significantly. If you use AI in recruitment, read our guide to fair AI resume screening; for lending, see how lenders use AI in underwriting responsibly.
What should a non-EU company selling into the EU check?
If you're based in India, the US or elsewhere, ask yourself these questions:
- Do EU residents use your AI agent? A chatbot on a website that serves EU customers, a voice agent answering calls from EU numbers, or a SaaS product with EU users all count as potential exposure.
- Are your agent's outputs used in the EU? For example, a candidate-ranking tool you run in India whose rankings are used by an EU client's hiring team.
- Are you the provider of an AI system that EU customers use? If you sell AI features in your product to EU businesses, you may have provider obligations, and your customers will start asking for documentation to meet their deployer obligations.
- Do you need an EU representative? Some non-EU providers of certain systems may need an authorized representative in the EU.
- What do your contracts say? EU clients are increasingly adding AI clauses about transparency, documentation and human oversight. Read them carefully before signing.
For businesses with no EU customers, no EU users and no EU clients, the Act may not apply directly, but it's still shaping expectations. Other jurisdictions are developing their own AI rules, and large customers often apply EU-style requirements to all their suppliers.
What do transparency duties mean for customer-facing agents?
For most businesses, this is the part of the Act that touches daily operations. In broad terms: when people interact directly with an AI system, they should be informed they're dealing with AI unless it's obvious from the context, and certain AI-generated content needs to be identifiable as such.
In practice, for a support or sales agent:
- Open conversations with a short, plain disclosure.
- Don't design the agent to claim it's human, even if asked.
- Make it easy to reach a person.
- Label AI-generated images, audio or video where the rules require.
Disclosure done well doesn't hurt the customer experience. Our guide on whether to tell customers they're talking to AI covers wording and design that keeps trust high.
What documentation habits should you build now?
Whatever tier your agents fall into, these habits make compliance easier and are good governance anyway:
- Keep an AI inventory. List every AI system and agent in use: what it does, who owns it, which vendor or model it uses, who it affects, and where those people are.
- Classify each use case against the risk tiers, with a short note on your reasoning. Revisit when the use changes.
- Record your role (provider or deployer) for each system.
- Collect vendor documentation. Ask AI vendors for instructions for use, intended purpose, known limitations and how they're approaching the Act.
- Document human oversight. Who reviews what, when, and with what authority to override.
- Keep logs of agent decisions and actions, with retention periods that respect data protection rules.
- Train staff. The Act expects providers and deployers to take measures to ensure sufficient AI literacy among staff who operate AI systems. Keep a record of what training was given.
- Write your disclosure text once and use it consistently across channels.
An AI usage policy is a natural home for most of this, and our post on what responsible AI looks like in a real business covers the broader practices.
Frequently Asked Questions
Does the EU AI Act apply to companies outside Europe?
It can. The Act can reach providers and deployers outside the EU when their AI systems are placed on the EU market or their outputs are used in the EU. Whether it applies to you depends on your customers, users and how your systems are used, so check with qualified counsel.
Is a customer service chatbot high-risk under the EU AI Act?
Generally, a chatbot that answers questions and books appointments isn't in the high-risk category, but it does carry transparency obligations, mainly informing people they're interacting with AI. If the same agent starts making decisions about eligibility for essential services or credit, the assessment can change.
When do EU AI Act obligations apply?
The Act applies in phases, with different obligations taking effect at different times, and the timeline has been the subject of discussion and proposed changes. Check the official EU sources or ask counsel for the current timeline that applies to your use case rather than relying on secondary summaries.
What are the penalties under the EU AI Act?
The Act sets out fines that scale with the seriousness of the violation and the size of the company, with the highest levels reserved for prohibited practices. Exact amounts and how they apply to smaller companies are set out in the regulation, so check the current text with your advisers.
Do I need a lawyer to comply with the EU AI Act?
For minimal-risk and transparency-tier uses, many businesses can handle the basics with good internal policies. For anything that may be high-risk, or if you're a provider selling AI systems into the EU, specialist legal advice is worth it. This post is a starting point, not a substitute.
Want to know where your AI agents stand?
If you're building or running AI agents that touch EU customers, tell us what your agents do and who they serve. We'll help you map each use case, build disclosure, logging and human oversight into the system, and prepare the documentation your counsel will ask for. See how we build compliant agents on our agentic AI development page.