EU AI Act compliance checker for online stores
Match your store’s AI uses to EU AI Act risk levels and duties. Your entries stay in this browser.
EU AI Act screening
Answer the questions
Complete the remaining questions for each use.
0 of 1 uses screened
Results by AI use
AI use 1
Answer the questions
Choose the AI use and its EU connection.
Screening of your answers. This does not certify compliance.
Answer the questions
The questions the checker asks
The EU AI Act compliance checker screens each AI use separately. It checks the EU connection, your role and the intended use before showing risk levels and duties.
Answer for your business and this particular use. Provider placed AI in EU covers putting the system on the EU market or into service there. Deployer based in EU covers operating the system when your business is established or located there. If both apply, select Both placement and location.
Output actually used in the EU can bring either role within scope. If you only operate a system in Australia and its output stays there, a vendor’s EU sales alone do not bring your use within scope. Choose No EU connection if none applies to your business for that use. Choose Unsure if you need to check.
These scope rules come from Article 2 of the AI Act. The checker covers commercial retail uses, not personal activities or every statutory exclusion.
Add a separate entry for each purpose, such as a shopper chatbot and an image generator. Choose Both roles if you provide the system and also operate it. The checker assigns duties only for the roles your EU connection covers.
You can be a provider if you develop or commission a system and put it into service under your name or trademark, including for your own use. You’re a deployer when you use AI under your authority. Rebranding, substantial modification or a change of intended purpose can change your responsibilities.
For an in-scope use, answer the restricted-practice question when it appears. Read these conditions before choosing None of the listed practices. If several restricted practices apply, record the use again for each one.
- Harmful manipulation covers subliminal, manipulative or deceptive techniques that impair informed decisions and cause or are likely to cause significant harm.
- Vulnerability abuse exploits age, disability or a social or economic vulnerability to distort behaviour with that harmful effect.
- Harmful social scoring assesses people over time and leads to detrimental treatment in unrelated contexts or treatment disproportionate to their behaviour.
- Sensitive biometric inference deduces race, political opinions, union membership, beliefs, sex life or sexual orientation from biometric data.
- Untargeted face scraping creates or expands facial-recognition databases from internet or CCTV images. Crime prediction uses profiling or personality alone.
- Staff or student emotion inference has a medical or safety exception. Live police face matching has narrow statutory authorisation conditions.
- Non-consensual intimate AI and child sexual abuse material have additional Article 5 conditions. Their new AI Act prohibitions start on 2 Dec 2026.
A flagged practice produces a stop-and-review result. The checker does not decide an exception or authorise the activity. Article 5 sets the full conditions.
Choose an Annex I section if the AI is a product or safety component covered by that section and the product requires a third-party health-and-safety conformity assessment. If you’re unsure, choose Safety AI, section unsure. Annex I lists the laws in each section.
For public-interest text, the follow-up asks whether human review or editorial control exists and someone holds editorial responsibility. Ordinary product descriptions belong under Ordinary AI text or images.
For a provider of generative AI, the first-market question tests the narrow machine-marking grace period. It does not extend chatbot notices or deployer disclosure deadlines.
Reading your result
Each entry has its own result. A use outside reported EU scope has no duties assigned by the checker. If the connection is unclear, check the scope before applying any duties.
Limited risk identifies transparency duties. A chatbot provider must make AI interaction clear by the first interaction unless it is obvious in context. If you use another provider’s chatbot, keep the notice visible so shoppers can see it.
Providers of generative AI have machine-readable marking and detectability duties. Deployers must disclose deepfakes and qualifying public-interest AI text. Human review or editorial control, with editorial responsibility, removes the public-interest text disclosure duty. It leaves the provider’s marking duty in place where it applies.
A deepfake falsely appears to be an authentic depiction of existing people, objects, places, entities or events. Evidently artistic or fictional work retains an adapted disclosure requirement. Human editorial review does not remove the deepfake disclosure rule.
The Article 50 transparency rules apply from 2 Aug 2026. Providers of systems first placed on the market before that date have until 2 Dec 2026 for marking under Article 50(2).
That grace period covers marking and detection only. Standard editing and changes that do not substantially alter supplied data or its meaning have a marking exception.
High-risk review flags hiring decisions, personal creditworthiness and relevant biometric uses. Other Annex III uses include education, essential services, critical infrastructure, law enforcement and justice. Financial-fraud detection alone is excluded from the personal-creditworthiness category.
Remote biometric identification excludes one-to-one verification whose sole purpose is confirming that someone is who they claim to be.
Article 6 allows a narrow exception for certain Annex III tasks that pose no significant harm and do not materially influence decisions. Profiling people in an Annex III use remains high-risk. A provider must document an exception assessment.
Annex III covers biometric categorisation that infers sensitive or protected attributes. Other biometric categorisation can still require the deployer to inform exposed people under Article 50(3). The checker screens Article 5 restrictions before considering Annex III.
The high-risk duties identified here start on 2 Dec 2027 for Annex III systems and 2 Aug 2028 for systems covered by Annex I Section A. Article 113 sets those dates. A qualifying Section B product goes to Sector rules review because Article 2(2) limits which AI Act provisions apply directly. If the section is unclear, the checker still asks about restricted practices. A flagged practice calls for review before use while you establish whether the Section B exception applies.
High-risk deployers prepare oversight, monitoring and controlled logs. If you deploy high-risk AI to assess personal creditworthiness, Article 27 requires a fundamental rights impact assessment before deployment. Providers prepare risk management, documentation and the applicable conformity assessment. Existing systems can have separate transitional rules.
Minimal risk means the selected purpose triggers none of the higher categories screened here. Privacy, consumer and product-safety laws still apply. Providers and deployers must support staff AI literacy for their operating context.
Load the hypothetical example for an Australian shop using another provider’s chatbot with EU customers and no restricted practice. It shows Transparency duties and a Limited risk result for that use.
Add Hiring or worker decisions as a separate Australian-only use. Choose No EU connection for that row. It shows Outside reported EU scope while the chatbot keeps its transparency result.
Download the JSON summary to retain the answers and each use’s result. Reloading clears your entries. The checker is a screening aid, not a legal opinion or compliance certificate.
General-purpose model-provider, importer and distributor obligations require a separate assessment. Use the EU AI Act guide to place these results in the wider rules.