Sources reviewed October 6, 2026. This explainer replaces the earlier broad country comparison. It is a research checklist, not an opinion on whether a particular business complies with the law.
Identify the activity before comparing countries
Write down what the system does, who uses it and which decisions it influences. A service that generates a draft and a system used in a regulated decision need different questions. Establish where the service is offered, where the affected people are and what role your organisation has. The vendor’s description is useful input, but it cannot answer every obligation for the deployer.
For example, a team adding a support chatbot should first document how users encounter it and what information it receives. A team considering a recruitment tool should describe the actual decision process and the people affected. Treat these as different research tasks; a shared model name does not make their obligations identical.
European Union: use the current implementation timeline
The European Commission’s AI Act overview describes staged application: prohibited-practice and AI-literacy provisions from February 2025, general-purpose-model obligations from August 2025 and wider application from August 2026, with exceptions. Its current timeline places specified high-risk use cases in December 2027 and product-embedded high-risk systems in August 2028. Read the relevant provisions and transition arrangements rather than treating August 2026 as the start date for every obligation.
The Commission’s enforcement overview also distinguishes authorities and provisions. This official guidance is informative; it does not replace the legislation or a case-specific assessment.
United States: voluntary guidance is not the whole picture
The FTC’s published remarks explain that using AI does not exempt a business from existing laws. Consumer protection, privacy and competition questions cannot be dismissed because a framework is voluntary. This source is dated February 2024; it establishes that distinction, not an exhaustive statement of every current federal or state rule.
For another jurisdiction, identify the actual authority and provision. Do not infer that another country has the same dates, definitions or exemptions because it uses similar terms such as risk, transparency or registration.
A rule-to-use worksheet
| Question | Evidence to keep |
|---|---|
| What activity is being assessed? | Actual workflow, affected people and the decisions involved |
| What rule is relevant? | Authority, provision, jurisdiction and publication version |
| When does it apply? | Application date, transition and any stated exception |
| Who must act? | The role defined by the rule and your role in the workflow |
| What is still unknown? | Questions requiring the provider, regulator or professional advice |
Keep uncertainty visible
A policy announcement, a political agreement, a final law and an enforcement notice are different documents. Keep each labelled with its status and date. Update the worksheet when the actual text or use changes, not merely when a new headline appears. A source list proves where information came from; it does not certify your compliance.
The useful outcome is a smaller list of questions tied to the real service. Keep unverified dates and interpretations out of a public promise. Where a decision has legal consequences, obtain advice for the particular jurisdiction and activity.
Review scope — October 10, 2026: the EU timeline was rechecked against the Commission’s current AI Act page. Dated FTC and policy material remains historical context. This article is a jurisdiction-specific reading guide, not a determination that a particular product or organisation complies with all applicable rules.
Compare two workflows before assigning a rule
Consider two fictional uses of the same model. In A, a support agent edits a draft answer before sending it. In B, a service ranks job applicants for a hiring team. Record the actual process separately; this example does not decide the risk class or legal obligations of either product.
| Record | Support draft A | Recruitment ranking B |
|---|---|---|
| Output | Editable reply | Applicant ranking |
| Affected people | Customers receiving a reply | People seeking employment |
| Decision path | Who edits and sends the answer? | Who uses the ranking and how? |
| Evidence gap | How users encounter AI and what data enters it | Purpose, role, affected people and applicable provisions |
For each workflow, copy the relevant authority, document version, provision, application date and any exception into a separate row. A vendor saying its model is compliant is not the same as checking your use. If the deployment region or the decision process is unknown, retain an unresolved question rather than choosing a convenient label.
Download the blank AI-use evidence sheet (CSV). It retrieves no law or vendor data and does not produce a compliance verdict. Keep customer, employee and applicant identifiers out of a sheet intended for sharing. Review the entries when the workflow, model supplier or applicable text changes.
Scoped source check, October 11, 2026: the Commission's current AI Act overview supports the staged timeline described above. The dated FTC material remains historical context; this check is not a complete review of every jurisdiction or a legal opinion.