Is my business a high-risk case?

In everyday language “high-risk AI” is almost any software that influences a decision about a person. In the law it is a technical term with a precise definition, and that definition does not fit most of the tools used by a trade business, a medical practice or a law firm. A chatbot on the website, AI-supported bookkeeping or text suggestions in the newsletter program do not, as a rule, fall under it. Anyone who wants to know for certain has to read two provisions: Article 6 and Annex III.

Two routes to high-risk

Article 6 has two separate routes by which an AI system counts as high-risk. The first concerns systems that serve as a safety component in a product which is already subject to a conformity assessment under existing EU product law, such as medical devices or machinery (paragraph 1, Annex I). That concerns manufacturers, not the businesses that merely use such products.

The second route is the relevant one for most Cautela customers: the closed list in Annex III (paragraph 2). If a use case is not on it, it is not a high-risk case by this route, however important the decision the system makes may feel.

The eight areas in Annex III

Annex III lists eight areas. Each is drawn more narrowly than its heading suggests:

• Biometrics: remote identification, categorisation by sensitive characteristics, emotion recognition.
• Critical infrastructure: safety components in road traffic and in the supply of water, gas, heating or electricity.
• Education: admission to educational institutions, evaluation of learning outcomes, monitoring of examinations.
• Employment: selection of applicants, decisions on promotion, termination or task allocation, performance evaluation.
• Access to essential services: public assistance benefits, creditworthiness assessment, risk assessment in life and health insurance, prioritisation of emergency calls.
• Law enforcement: risk assessment, lie detection, evaluation of evidence.
• Migration, asylum and border control.
• Administration of justice and democratic processes: assisting courts, influencing elections.

For the typical Cautela customer, trades, practices, law firms, retail, agencies, two areas are the most likely to matter in practice: selection of applicants (where an AI tool filters or rates applications) and creditworthiness assessment (where a business itself makes credit decisions with AI support, for instance for payment by instalments). The other six areas mostly concern public authorities, financial institutions or regulated sectors.

The derogation that is often overlooked

Finding a matching point in Annex III does not automatically put a system in the high-risk category. Article 6(3) takes a whole series of cases back out:

“… an AI system referred to in Annex III shall not be considered to be high-risk 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.” Regulation (EU) 2024/1689, Article 6(3)

In concrete terms this applies if one of four things is true: the system performs a narrow procedural task. It improves the result of a human activity that has already been completed. It detects deviations from earlier decision-making patterns without replacing the human review. Or it merely prepares an assessment that someone else makes. Software that pre-sorts application documents by formal criteria, while a person always decides in the end, can drop out of the high-risk category in this way even though it sits in the “selection of applicants” field.

There is one express limit: as soon as a system performs profiling of natural persons, that is, automatically evaluates a person's characteristics, behaviour or preferences, it always counts as high-risk regardless of the four exceptions.

A provider who considers that an Annex III system is not high-risk in its own case has to document that assessment before the system is placed on the market and present it to the authority on request (Article 6(4)).

Deadlines

The duties for systems that really are high-risk do not bite immediately. The Digital Omnibus moved the deadlines in the summer of 2026: for systems under Annex III, Chapter III applies from 2 December 2027, and for high-risk AI in already regulated products under Annex I from 2 August 2028 (Article 113 as amended).

What to do where use really is high-risk

If a business does operate a genuine Annex III system that falls under none of the exceptions, Article 26 applies. The most important duties under it: use the system in accordance with the instructions for use, assign human oversight to persons who have the necessary competence and training, monitor operation continuously and inform the provider and the authority of risks, keep automatically generated logs for at least six months, and inform workers' representatives before the system is used in the workplace. Breaches are subject to fines of up to €15,000,000 or 3 % of worldwide annual turnover (Article 99(4)(e)).

Cautela itself falls under none of the eight Annex III areas and is therefore not a high-risk system. What remains for Cautela, as for most customers, is the transparency obligation under Article 50, because a conversation with an AI system takes place. There is a separate article on that.