Practical requirements extracted from policy instruments, with the source article, the actors they bind, evidence examples and original framework mappings. Legal requirements are marked; everything else is voluntary guidance.
High-risk AI systems must achieve an appropriate level of accuracy, robustness and cybersecurity and perform consistently throughout their lifecycle. Accuracy levels and metrics must be declared in the instructions; systems must be resilient to errors, faults and inconsistencies, address feedback loops in continuously learning systems, and resist attempts to alter use or performance, including data poisoning, model poisoning, adversarial examples and confidentiality attacks.
Pending review (a factual check against the official source, not a legal review)Applies from 2 Dec 2027
Legal requirementConformity assessment and registration·European Union
EU AI Act · Articles 43, 47, 48 and 49; Annex VIII
Before placing a high-risk system on the market, providers must complete the applicable conformity assessment (internal control or notified-body assessment depending on the system), draw up an EU declaration of conformity, affix the CE marking, and register the system in the EU database. Deployers that are public authorities must also register their use of Annex III systems.
Pending review (a factual check against the official source, not a legal review)Applies from 2 Dec 2027
Legal requirementData governance and quality·European Union
High-risk AI systems that use data-driven techniques must be developed on training, validation and testing data sets meeting quality criteria: appropriate governance practices covering design choices, data collection and origin, preparation, assumptions, availability and suitability, examination for possible biases, and measures to detect, prevent and mitigate bias. Data must be relevant, sufficiently representative and, to the best extent possible, free of errors and complete for the intended purpose.
Pending review (a factual check against the official source, not a legal review)Applies from 2 Dec 2027
Legal requirementGovernance and accountability·European Union
A deployer using a high-risk post-remote biometric identification system in a criminal investigation must request authorisation from a judicial or independent administrative authority in advance or within 48 hours, unless the use is only the initial identification of a potential suspect on objective and verifiable facts. Use is limited to a targeted search; if authorisation is refused, use stops and the data is deleted. No adverse legal decision may rest solely on the output, each use is documented, and the deployer reports annually to the market surveillance and data protection authorities.
Verified (a factual check against the official source, not a legal review)Applies from 2 Dec 2027
Legal requirementGovernance and accountability·European Union
Before making a high-risk AI system available in the Union, a provider established outside the EU must appoint, by written mandate, an authorised representative established in the Union. The representative verifies that the EU declaration of conformity and technical documentation exist and that conformity assessment was done, keeps the provider's contact details and the documentation available for ten years, supplies information to authorities, cooperates with them, and must end the mandate if the provider acts contrary to the Regulation.
Verified (a factual check against the official source, not a legal review)Applies from 2 Dec 2027
Legal requirementGovernance and accountability·European Union
Article 16 lists what a provider of a high-risk AI system owes: compliance with the Section 2 requirements, its name and contact details on the system or its documentation, a quality management system, retention of documentation and logs, conformity assessment, an EU declaration of conformity, CE marking, registration, corrective action when needed, cooperation with authorities on request, and accessibility in line with the EU accessibility directives.
Verified (a factual check against the official source, not a legal review)Applies from 2 Dec 2027
Legal requirementGovernance and accountability·European Union
On a reasoned request from a national competent authority, a provider of a high-risk AI system must supply all the information and documentation needed to show that the system meets the Section 2 requirements, in a language the authority can readily understand, and must give the authority access to the automatically generated logs it holds. Authorities must treat what they receive as confidential under Article 78.
Verified (a factual check against the official source, not a legal review)Applies from 2 Dec 2027
EU AI Act · Article 14; Article 26(2) for deployers
High-risk systems must be designed with human-machine interface tools so natural persons can effectively oversee them, understand capacities and limitations, avoid automation bias, interpret output, decide not to use the system, and intervene or stop it. Deployers must assign oversight to people with the necessary competence, training and authority. For certain remote biometric identification systems, action requires verification by at least two competent persons.
Pending review (a factual check against the official source, not a legal review)Applies from 2 Dec 2027
ICO AI guidance · UK GDPR Article 35; ICO guidance, accountability and governance section
Where AI processing of personal data is likely to result in a high risk to individuals, UK GDPR requires a DPIA before processing begins. The ICO treats most AI involving profiling, large-scale processing or novel technology as meeting this threshold.
Source-linked (a factual check against the official source, not a legal review)
UAE PDPL · Article on data protection impact assessment (reviewer to cite article number)
Before processing that uses modern technologies and is likely to pose a high risk to privacy, controllers must assess the impact on personal data protection, covering the processing, its purposes, risks and safeguards.
Pending review (a factual check against the official source, not a legal review)
Legal requirementIncident reporting and handling·European Union
Providers of high-risk AI systems must report serious incidents to the market-surveillance authority of the Member State where the incident occurred, immediately after establishing a causal link (or reasonable likelihood) and no later than 15 days after becoming aware, with shorter limits for the most serious cases such as widespread infringements or death. Deployers must inform the provider and authorities when they identify a serious incident.
Pending review (a factual check against the official source, not a legal review)Applies from 2 Dec 2027
Legal requirementPrivacy and personal-data protection·Nepal
Nepal Privacy Act 2075 · Chapter on collection and protection of personal information (reviewer to cite sections)
Personal information may be collected only by authorised persons for a lawful purpose with the individual's consent, and must not be used or disclosed for other purposes without consent, subject to statutory exceptions. AI systems trained on or processing personal data of people in Nepal must respect these limits.
Pending review (a factual check against the official source, not a legal review)
Legal requirementPrivacy and personal-data protection·European Union
Where a deployer of a high-risk AI system is required to carry out a data protection impact assessment under Article 35 of the GDPR or Article 27 of the Law Enforcement Directive, it must draw on the information the provider supplied under Article 13, such as the intended purpose, performance, limitations and human-oversight measures, when producing that assessment.
Verified (a factual check against the official source, not a legal review)Applies from 2 Dec 2027
Article 5 bans placing on the market, putting into service or using AI for listed practices, including subliminal or manipulative techniques that cause significant harm, exploitation of vulnerabilities, social scoring by public or private actors leading to detrimental treatment, untargeted scraping of facial images to build recognition databases, emotion recognition in workplaces and education institutions except for medical or safety reasons, biometric categorisation to infer protected characteristics, and real-time remote biometric identification in publicly accessible spaces for law enforcement outside narrow exceptions.
Pending review (a factual check against the official source, not a legal review)Applies from 2 Feb 2025
A governmental entity may not develop or deploy an AI system for the purpose of uniquely identifying a specific individual using biometric data, or of gathering images or other media from the internet or another public source without the individual's consent, where doing so would infringe a right guaranteed under the United States or Texas constitutions or violate state or federal law. The Act preserves lawful uses that comply with existing biometric-privacy law.
Verified (a factual check against the official source, not a legal review)Applies from 1 Jan 2026
Legal requirementPublic-sector use and procurement·European Union
EU AI Act · Article 26(8); Article 49(3) and 49(4)
Deployers that are public authorities or Union institutions, bodies, offices or agencies must register their use of an Annex III high-risk AI system in the EU database before putting it into service. If they find that the system they intend to use has not been registered in the database by its provider, they must not use it and must inform the provider or distributor.
Verified (a factual check against the official source, not a legal review)Applies from 2 Dec 2027
Legal requirementRecord keeping and logging·European Union
EU AI Act · Article 12; Article 26(6) for deployers
High-risk AI systems must technically allow automatic recording of events (logs) over their lifetime to support traceability, post-market monitoring and operational monitoring. Deployers must keep the logs generated by the system, to the extent under their control, for a period appropriate to the intended purpose and at least six months unless other law provides otherwise.
Pending review (a factual check against the official source, not a legal review)Applies from 2 Dec 2027
Legal requirementRecord keeping and logging·European Union
Providers must keep the event logs that a high-risk AI system generates under Article 12, to the extent those logs are within their control, for a period appropriate to the system's intended purpose and in any case for at least six months, unless Union or national law on personal data sets a different period. Financial institutions keep the logs under their sector rules.
Verified (a factual check against the official source, not a legal review)Applies from 2 Dec 2027
An AI business operator that provides high-impact AI, or a product or service using it, must establish and operate a plan for managing the risks of that AI. High-impact AI is AI used in areas such as energy, drinking water, health care and medical devices, nuclear safety, biometric analysis for criminal investigation, decisions on recruitment or loans, transport, public services and education, where it may materially affect life, safety or fundamental rights.
Verified (a factual check against the official source, not a legal review)Applies from 22 Jan 2026
Legal requirementTransparency and disclosure·European Union
A deployer of an emotion recognition system or a biometric categorisation system must inform the natural persons exposed to it that the system is operating and must process their personal data in line with the GDPR, the Law Enforcement Directive and the EU institutions data protection regulation. Systems permitted by law to detect, prevent or investigate criminal offences are excepted, subject to safeguards. Use in workplaces and schools is separately banned by Article 5 except for medical or safety reasons.
Verified (a factual check against the official source, not a legal review)Applies from 2 Aug 2026
Voluntary guidanceSafety testing and evaluation·United States
Measure covers selecting metrics and test methods, evaluating validity, safety, security, resilience, explainability, privacy, fairness and bias, and monitoring these over time, including through independent review and red-teaming for generative AI.
Source-linked (a factual check against the official source, not a legal review)
A single practical requirement pulled out of an instrument and stated on its own: keep a risk management system, log incidents, document training data, provide human oversight, and so on. Each one cites the article or section it comes from so you can check it against the source.
Does a voluntary obligation have legal force?
No, and every obligation is labelled either a legal requirement or voluntary guidance. Voluntary items still matter in practice, because procurement questionnaires and auditors ask about them, but only the binding ones carry legal consequence.
How do I find the obligations that apply to my organisation?
Filter by jurisdiction, category, actor, sector or use case. The applicability check asks a short set of questions and returns the duties that may reach you. It is an educational screen, not a legal determination, and it says so.
Why do some instruments have no obligations listed?
Because nobody has broken them out yet. Most instruments are recorded at summary level first; obligations are added jurisdiction by jurisdiction. The coverage and open-gaps pages publish exactly what is missing rather than hiding it.
