Is our LLM use case high-risk under the EU AI Act? Annex III for companies using AI APIs
Last updated
Under the AI Act, Regulation (EU) 2024/1689, a model is not high-risk because it is powerful. A use is high-risk because of what it is used for. Article 6(2) makes every AI system listed in Annex III high-risk, Article 6(3) lets a narrow set of them out, and profiling of people keeps them in. The rules for Annex III systems apply from 2 December 2027, after Regulation (EU) 2026/1744 moved the date.
This page walks through the test for a company that uses ChatGPT, Claude, Copilot or LLM APIs, classifies eight common uses, and lists what changes for you as a deployer when a use is high-risk. Quotes are from the Official Journal, checked on 9 November 2026. It is not legal advice.
Create a company workspaceor write to contact@standardcompute.com
DeutschFrançaisNederlandsEspañolItalianoSvenskaNorskDanskSuomiPolski
Short answers
- Is using ChatGPT, Claude or an LLM API high-risk under the EU AI Act?
- Not by itself. High-risk status follows the use. An internal coding assistant, a drafting tool or a support chatbot is not listed in Annex III. The same model becomes part of a high-risk system when you use it to screen CVs, evaluate employees, score credit, price life or health insurance, or grade students.
- Which LLM uses are high-risk under Annex III?
- For most companies the relevant points are 4(a) recruitment and selection, including filtering applications and evaluating candidates; 4(b) decisions on work terms, promotion or termination, task allocation based on behaviour or traits, and monitoring or evaluating workers; 5(b) creditworthiness and credit scores of natural persons; 5(c) risk assessment and pricing in life and health insurance; and 3 education, such as evaluating learning outcomes.
- Can an Annex III use escape high-risk status?
- Yes, under Article 6(3), if the system does not pose a significant risk of harm and meets one of four conditions, such as a narrow procedural task or a preparatory task. The provider documents that assessment and registers the system under Article 49(2). The derogation never applies where the system performs profiling of natural persons.
- When do the high-risk rules apply?
- From 2 December 2027 for Annex III systems and from 2 August 2028 for AI systems in products covered by Annex I, under Article 113 as amended by Regulation (EU) 2026/1744. AI literacy under Article 4 has applied since 2 February 2025.
The test in four questions
Work through the questions for each use. The same LLM can sit in a low-risk use in one team and a high-risk use in another.
| Question | Where in the AI Act | If yes |
|---|---|---|
| 1. Is the use in one of the eight Annex III areas? Biometrics, critical infrastructure, education, employment, essential private and public services, law enforcement, migration and border control, justice and democratic processes. | Where in the AI Act: Article 6(2), Annex III points 1 to 8 | If yes: Go to question 2. If no, the use is not high-risk under Annex III. |
| 2. Does it match a specific point, such as filtering job applications (4(a)) or establishing a credit score (5(b))? | Where in the AI Act: Annex III, the lettered points | If yes: Go to question 3. A use that is only near an area is not listed. |
| 3. Does the system profile natural persons, meaning it evaluates personal aspects such as work performance, economic situation, behaviour or reliability? | Where in the AI Act: Article 6(3), third subparagraph; Article 3(52) and GDPR Article 4(4) | If yes: It is high-risk. The derogation is not available. |
| 4. Does it not materially influence the outcome and meet one of the four conditions in Article 6(3)? | Where in the AI Act: Article 6(3), first and second subparagraphs; Article 6(4) | If yes: It may be outside high-risk. The provider documents the assessment and registers it under Article 49(2). |
Summary of Regulation (EU) 2024/1689 as amended by Regulation (EU) 2026/1744, checked on 9 November 2026. Not legal advice.
Common LLM uses, classified
The table is our reading of Annex III for typical uses of LLM APIs. The deciding factor is the intended purpose, defined in Article 3(12) as the use for which a system is intended, including the specific context and conditions of use.
| Use | Annex III | Our reading |
|---|---|---|
| Internal coding assistant (Claude Code, Copilot, an agent on an LLM API) | Annex III: Not listed | Our reading: Not high-risk. Article 4 AI literacy applies to the people using it. |
| Customer support chat | Annex III: Not listed as such | Our reading: Not high-risk while it answers questions. If you build it, you are its provider and Article 50(1) requires telling people they are talking to an AI system, unless that is obvious. It moves into Annex III if it decides on credit (5(b)) or public benefits (5(a)). |
| Drafting, summarising and translating documents | Annex III: Not listed | Our reading: Not high-risk for ordinary business documents. It needs a second look when the output evaluates people in an Annex III area, such as a draft performance review. |
| CV screening, ranking or filtering candidates | Annex III: Point 4(a) | Our reading: High-risk. Evaluating candidates is profiling, so the Article 6(3) derogation is not available. |
| Employee monitoring, performance evaluation, task allocation based on behaviour | Annex III: Point 4(b) | Our reading: High-risk where the system allocates tasks based on individual behaviour or personal traits or monitors and evaluates performance and behaviour. |
| Credit scoring of natural persons | Annex III: Point 5(b) | Our reading: High-risk. Systems used to detect financial fraud are excepted. A fundamental rights impact assessment is required (Article 27). |
| Risk assessment and pricing for life and health insurance | Annex III: Point 5(c) | Our reading: High-risk. Motor, property and other non-life pricing is not in point 5(c). A fundamental rights impact assessment is required (Article 27). |
| Grading exams or evaluating learning outcomes | Annex III: Point 3(b) | Our reading: High-risk in educational and vocational training institutions at all levels, including when the results steer the learning process. |
Our reading of Annex III, checked on 9 November 2026. Your own facts decide the classification. Not legal advice.
Annex III point 4(a) reads: “AI systems intended to be used for the recruitment or selection of natural persons, in particular to place targeted job advertisements, to analyse and filter job applications, and to evaluate candidates”.
Point 4(b) reads: “AI systems intended to be used to make decisions affecting terms of work-related relationships, the promotion or termination of work-related contractual relationships, to allocate tasks based on individual behaviour or personal traits or characteristics or to monitor and evaluate the performance and behaviour of persons in such relationships.”
Point 5(b) reads: “AI systems intended to be used to evaluate the creditworthiness of natural persons or establish their credit score, with the exception of AI systems used for the purpose of detecting financial fraud”. Point 5(c) reads: “AI systems intended to be used for risk assessment and pricing in relation to natural persons in the case of life and health insurance”.
Article 6(3): the derogation and the profiling exception
Article 6(3) takes an Annex III system out of high-risk status where it does not pose a significant risk of harm to the health, safety or fundamental rights of natural persons, including by not materially influencing the outcome of decision making. One of four conditions must be met: the system performs a narrow procedural task; it improves the result of a previously completed human activity; it detects decision-making patterns or deviations without replacing or influencing the human assessment without proper human review; or it performs a preparatory task to an assessment relevant to an Annex III use.
The third subparagraph reads: “Notwithstanding the first subparagraph, an AI system referred to in Annex III shall always be considered to be high-risk where the AI system performs profiling of natural persons.” A tool that ranks candidates or scores employees evaluates people and cannot use the derogation.
Under Article 6(4) the provider makes this call, documents it before the system is placed on the market or put into service, and hands the documentation to national authorities on request. Under Article 49(2) that provider also registers itself and the system in the EU database. Regulation (EU) 2026/1744 kept that registration and shortened the information required by deleting points 7 and 9 of Annex VIII, Section B. It did not amend Article 6(3), Article 6(4) or Annex III.
Article 6(5) asked the Commission for guidelines with practical examples by 2 February 2026. The Commission published draft guidelines on 19 May 2026 for consultation. As at 9 November 2026 no final version had been adopted, so this page does not rely on the draft. Under Article 7, the Commission can add or modify Annex III use cases by delegated act.
When your company becomes the provider
Most companies using AI are deployers: they use an AI system under their own authority (Article 3(4)). A company that builds its own CV screener or scoring tool on top of an LLM API and puts it into service under its own name is the provider of that system (Article 3(3)), with the provider duties for high-risk systems, including the conformity assessment.
Article 25(1) adds three cases in which a deployer becomes the provider of a high-risk system: it puts its name or trademark on one; it makes a substantial modification so that it remains high-risk; or it modifies the intended purpose of an AI system, “including a general-purpose AI system, which has not been classified as high-risk”, so that it becomes high-risk. Pointing a general-purpose assistant at job applications is the case to watch.
Regulation (EU) 2026/1744 replaced Article 25(2): the initial provider must cooperate with the new provider, including by making technical documentation available, informing it about known limitations and failure modes, and giving targeted technical access for testing, unless it clearly specified that its system is not to be changed into a high-risk system.
What changes for a deployer when a use is high-risk
These duties apply to Annex III systems from 2 December 2027. Article 26 was not amended by Regulation (EU) 2026/1744.
| Duty | Provision | What to do |
|---|---|---|
| Use the system according to the instructions for use | Provision: Article 26(1) | What to do: Get the provider's instructions for use and build your process around them |
| Human oversight | Provision: Article 26(2) | What to do: Name the people who oversee the system and give them the competence, training, authority and support |
| Relevant and representative input data | Provision: Article 26(4) | What to do: Where you control the input data, check it fits the intended purpose |
| Monitor, report and suspend | Provision: Article 26(5) | What to do: Monitor operation, inform the provider and the market surveillance authority of risks, suspend use where needed, report serious incidents |
| Keep the logs | Provision: Article 26(6) | What to do: Keep the logs the high-risk system generates automatically, where under your control, for at least six months |
| Inform workers | Provision: Article 26(7) | What to do: Before using a high-risk system at the workplace, inform workers' representatives and the affected workers |
| Inform affected people | Provision: Article 26(11) | What to do: Tell people that decisions about them are made or assisted by a high-risk system |
| Data protection impact assessment | Provision: Article 26(9) | What to do: Use the provider's information under Article 13 for your DPIA under GDPR Article 35 |
| Fundamental rights impact assessment | Provision: Article 27 | What to do: Required for public bodies, private entities providing public services, and deployers of credit-scoring (5(b)) and life and health insurance (5(c)) systems; it can cross-reference your DPIA |
| Explain decisions on request | Provision: Article 86(1) | What to do: Give affected people clear and meaningful explanations of the role of the AI system in decisions with legal or similarly significant effects |
Summary of Articles 26, 27 and 86 of Regulation (EU) 2024/1689, checked on 9 November 2026. Not legal advice.
The dates
| Date | What applies | Provision |
|---|---|---|
| 2 February 2025 | What applies: Chapters I and II, including AI literacy (Article 4) and the prohibited practices | Provision: Article 113(3)(a) |
| 2 August 2026 | What applies: General date of application, including the transparency duties in Article 50 | Provision: Article 113 |
| 2 December 2027 | What applies: High-risk rules for Annex III systems, including Articles 26 and 27 | Provision: Article 113(3)(c)(i), as amended by Regulation (EU) 2026/1744 |
| 2 August 2028 | What applies: High-risk rules for AI systems in products covered by Annex I | Provision: Article 113(3)(c)(ii), as amended by Regulation (EU) 2026/1744 |
| 2 August 2030 | What applies: High-risk systems used by public authorities, including systems already in place | Provision: Article 111(2), as replaced by Regulation (EU) 2026/1744 |
From the Official Journal text, checked on 9 November 2026. Not legal advice.
What else Regulation (EU) 2026/1744 changed, including the two new prohibited practices from 2 December 2026, is on the page about the AI Act digital omnibus.
Checklist for companies using LLMs
| Step | What to do | Where to start |
|---|---|---|
| 1. List every use | What to do: One line per use: team, purpose, model or vendor, who is affected, whether people are evaluated | Where to start: EU AI Act deployer checklist |
| 2. Classify each use | What to do: Run the four questions above and record the answer, the Annex III point and the date | Where to start: This page |
| 3. Flag profiling | What to do: Mark every use that evaluates candidates, employees, customers or students | Where to start: Article 6(3), third subparagraph |
| 4. Check whether you are the provider | What to do: Own tools built on an LLM API and any change of intended purpose under Article 25(1) | Where to start: EU AI Act obligations |
| 5. Plan the deployer duties | What to do: Oversight owners, log retention, worker information, notices to affected people, DPIA and, where required, FRIA | Where to start: The table above |
| 6. Ask your vendors | What to do: Intended purpose, instructions for use, logs, registration status and which models and providers process your data | Where to start: LLM gateway RFP template |
| 7. Train the people involved | What to do: Overseers of high-risk uses need more than the base module | Where to start: AI literacy under Article 4 |
| 8. Review on change | What to do: Re-run the test when a use, a vendor or the purpose changes, and when the Commission adopts its Article 6 guidelines | Where to start: Dated record of decisions |
Not legal advice. Your legal team decides what the AI Act requires of your company.
What a gateway does and does not do
A gateway does not make a high-risk use compliant. It does not classify your uses, provide human oversight, carry out a fundamental rights impact assessment or produce the logs a high-risk system must generate. Those stay with the provider and with you as deployer.
What it can do is keep the record of which models and providers each use runs on. Standard Compute is one option for model APIs, Claude Code and Claude Desktop: give every person and every app its own key, all under one company account, and decide which models each key can use. Choose exactly which inference providers may handle your data, for the whole company, a team or a single key.
Every admin change is recorded with before and after values, and you can export a record of every request showing the model, the provider that served it and the region rule in force, kept for 12 months. Request records hold metadata only, never your prompts or outputs.
Questions from legal, compliance and procurement
- Is a general-purpose model such as GPT or Claude a high-risk AI system?
- No. General-purpose AI models have their own rules in Chapter V of the AI Act, which bind the model provider. Whether a system built on such a model is high-risk depends on its intended purpose under Article 6 and Annex III.
- We only use an LLM to summarise CVs for a recruiter. Is that high-risk?
- Annex III point 4(a) covers systems used to analyse and filter job applications and to evaluate candidates. A summary that ranks, scores or filters candidates is inside it, and evaluating candidates is profiling, so Article 6(3) does not help. Record your reasoning either way.
- Did Regulation (EU) 2026/1744 change Annex III or Article 6(3)?
- No. It inserted Article 6(1a) to (1c) on safety components in products, moved the Annex III date to 2 December 2027, shortened the registration content in Annex VIII and replaced Articles 25(2) and 27(4). The list of high-risk uses and the derogation read as before.
- Are the Commission's guidelines on high-risk classification final?
- No. Article 6(5) required them by 2 February 2026. The Commission published a draft on 19 May 2026 for consultation, and no final version had been adopted as at 9 November 2026. The guidelines are not legally binding.
- Is fraud detection in lending high-risk?
- Point 5(b) excepts AI systems used for the purpose of detecting financial fraud. A system that also feeds the creditworthiness decision or the credit score of a natural person is inside point 5(b).
Related pages
- EU AI Act obligations for companies using LLMs
- Free EU AI Act deployer checklist
- AI literacy under the EU AI Act (Article 4)
- What the AI Act digital omnibus (2026/1744) changed
- Free LLM gateway RFP and tender requirements template
- Standard Compute for enterprise
Sources
- Regulation (EU) 2024/1689 (AI Act): Articles 3, 6, 7, 25, 26, 27, 49, 50, 86, 113 and Annex III (EUR-Lex), checked 2026-11-09
- Regulation (EU) 2026/1744 (Digital Omnibus on AI), OJ L, 24 July 2026: Article 1, points (8), (12), (13), (40) and (42) (EUR-Lex), checked 2026-11-09
- AI Act, Article 6: classification rules for high-risk AI systems (AI Act Service Desk), checked 2026-11-09
- AI Act, Annex III: high-risk AI systems referred to in Article 6(2) (AI Act Service Desk), checked 2026-11-09
- AI Act, Article 25: responsibilities along the AI value chain (AI Act Service Desk), checked 2026-11-09
- AI Act, Article 26: obligations of deployers of high-risk AI systems (AI Act Service Desk), checked 2026-11-09
- AI Act, Article 27: fundamental rights impact assessment (AI Act Service Desk), checked 2026-11-09
- AI Act, Article 49: registration (AI Act Service Desk), checked 2026-11-09
- AI Act, Article 113: entry into force and application (AI Act Service Desk), checked 2026-11-09
- European Commission: draft guidelines on the classification of high-risk AI systems, 19 May 2026, draft (digital-strategy.ec.europa.eu), checked 2026-11-09
Create a company workspace, give every person and app its own key with the models and providers you approve, and keep the admin record next to your classification of each use.