What Counts as a High-Risk AI System Under the EU AI Act? 

A high-risk AI system is one that meets one of the classification routes in Article 6 of the EU Artificial Intelligence Act (AI Act). “High risk” does not simply mean the artificial intelligence is powerful, complex, or used in a regulated industry. Classification turns on the system’s intended purpose, use context, and connection to specified products or sensitive use cases.

The EU AI Act provides two main routes: certain AI systems tied to products covered by legislation listed in Annex I and systems used for specified purposes listed in Annex III. High-risk AI systems are not prohibited systems. They may be placed on the EU market or put into service when they meet the applicable requirements. U.S.-based organizations may also need to assess artificial intelligence systems when the Act’s territorial scope applies.

Key takeaways:

• Start with Article 6 rather than the organization’s industry.
• Evaluate both the Annex I and Annex III routes.
• If a system is high risk, document the rationale, organizational role, and resulting obligations.

Article 6 Creates Two Routes to High-Risk Classification

An AI system is not high risk simply because it uses generative or agentic AI; classification depends on its intended purpose and the Article 6 criteria.

The first route covers certain AI systems connected to regulated products. An AI system can qualify when it is itself a product covered by the EU legislation listed in Annex I or is intended to function as a safety component of one. The product must also be subject to third-party conformity assessment under the applicable Annex I legislation for relevant health or safety requirements. Conformity assessment is the process used to determine whether applicable requirements have been met.

AI used solely for user assistance, efficiency, automation, convenience, or quality control does not qualify as a safety component on that basis alone. An AI system whose failure or malfunction could endanger health or safety can qualify as a safety component.

 The second route covers specified Annex III uses. Classification depends on what the provider intends the system to do and the context and conditions in which it is intended to operate.

Annex III Covers Specific Sensitive AI Use Cases

Annex III identifies high-risk use cases across eight broad areas. An AI application operating in one of these sectors is not automatically a high-risk system.

An AI system evaluating an employment candidate may fall within Annex III, while an administrative scheduling tool generally would not. A system determining a person’s creditworthiness also differs from software that only explains an application form. Whether the system is used in employment, law enforcement, critical infrastructure, or border control, the specific intended use determines whether the Annex III criteria apply.

Some Annex III Systems Can Fall Outside High-Risk Classification

Article 6(3) provides a limited filter for certain Annex III systems. A provider may conclude that a system is not high risk when it does not pose a significant risk of harm to health, safety, or fundamental rights, including because it does not materially influence decision-making, and at least one specified condition applies.

Those conditions cover systems that:

• Perform a narrow procedural task.
• Improve the result of a previously completed human activity.
• Detect decision-making patterns or deviations without replacing or influencing a previously completed human assessment, with appropriate human review.
• Perform a preparatory task for an assessment associated with an Annex III use case.

A system evaluating eligibility for essential services may be high risk, while a tool limited to structuring information for human assessment may qualify for the filter. An Annex III system that profiles natural persons cannot rely on it.

Once the relevant provisions apply, a provider relying on Article 6(3) must document the assessment before placing the system on the market or putting it into service and meet the applicable registration requirement. The European Commission has also published draft, nonbinding classification guidance with practical examples.

How To Assess Whether an AI System Is High Risk

Organizations can assess classification in five steps:

1. Document the system’s intended purpose, users, and context and conditions of use.
2. Evaluate the Annex I route, including whether the AI is a covered product or safety component and whether both Article 6(1) conditions apply.
3. Compare the intended use with the specific Annex III categories rather than the organization’s industry alone.
4. Apply the Article 6(3) filter where relevant, including the significant-risk test, specified conditions, and profiling exception.
5. Document the classification, supporting rationale, organizational role, and requirements that follow.

An organization should be able to trace its conclusion to the intended purpose, classification route, supporting evidence, and its role. That record gives AI governance and compliance teams a clearer basis for evaluating technical documentation and risk management controls. Revisit the analysis if the intended purpose or use changes.

High-Risk AI Compliance Checklist

Once an AI system is classified as high risk, the organization should identify which requirements apply based on the classification route and its role. For providers of Annex III high-risk systems, the review should address.

• Risk management: Is a risk management system in place?
• Data governance: Are applicable data governance requirements addressed?
• Technical documentation: Does the documentation describe the system and support the compliance rationale?
• Logging and recordkeeping: Are required logs and records available?
• Transparency and instructions: Are deployers given the information they need to use the system appropriately?
• Human oversight: Are qualified people able to monitor operation and intervene when appropriate?
• Accuracy, robustness, and cybersecurity: Are applicable performance and resilience requirements addressed?
• Quality management: Is the required quality management framework in place?
• Conformity assessment: Has the applicable conformity assessment pathway been identified?
• Registration and post-market monitoring: Are applicable registration and ongoing monitoring requirements addressed?

Deployers have a different set of responsibilities, which may include following instructions, assigning human oversight, monitoring relevant risks, maintaining applicable logs, and meeting notification or information requirements. Certain deployers must also complete a fundamental rights impact assessment before deploying specified high-risk systems. 

For Annex I systems, the applicable requirements and conformity pathway also depend on the relevant product legislation. Not every product-related high-risk system follows the same checklist.

 Under the current high-risk AI timeline, the high-risk regime applies beginning Dec. 2, 2027, to systems in specified Annex III areas and Aug. 2, 2028, to high-risk systems embedded in regulated products. Other EU AI Act provisions are already applicable.

High-Risk AI System FAQs

Are High-Risk AI Systems Prohibited Under the EU AI Act?

No. High-risk AI systems and prohibited AI practices are separate concepts. High-risk systems may be used when applicable requirements are satisfied.

Is Every AI System Listed in Annex III Automatically High Risk?

No. The Article 6(3) filter can apply when the system does not pose a significant risk to health, safety, or fundamental rights and at least one specified condition is met. Systems that profile natural persons cannot rely on that filter.

Does Every AI System Used in a Regulated Industry Count as High Risk?

No. Classification depends on intended purpose and the specific Article 6 or Annex III criteria, not the organization’s industry label. 

When Do the EU AI Act High-Risk Requirements Apply?

Rules for systems in specified Annex III high-risk areas apply beginning Dec. 2, 2027. Rules for high-risk systems embedded in regulated products apply beginning Aug. 2, 2028. Other provisions of the AI Act are already applicable.

Evaluate Your EU AI Act Readiness

Organizations preparing for the high-risk regime should be able to show what each AI system is intended to do, how it was classified, what evidence supports that conclusion, which role the organization holds, and which requirements follow. 

Insight Assurance can independently evaluate classification rationale, documentation, evidence, and control alignment as part of EU AI Act readiness. Insight Assurance does not make an organization’s underlying legal classification decision or design and operate its controls.

Last, but not least, remember that high-risk classification depends on intended purpose and use context. The same underlying AI technology may fall outside the high-risk framework in one use case and meet the criteria in another, so the classification analysis should focus on how the system is intended to be used.

Talk to an expert about your EU AI Act readiness.