What the AI Act is
The AI Act is an EU regulation that sorts AI systems by risk level: banned practices (Article 5), high-risk systems, systems with transparency duties (Article 50) and all others. As a regulation, it applies directly in every EU country, with no national laws needed to put it into effect.
The penalties are serious. For breaking the bans of Article 5, the fine is up to 35 million euros or up to 7% of worldwide annual turnover for the previous year. For breaking the duties of providers and deployers and the transparency rules of Article 50, it is up to 15 million euros or up to 3% of turnover, whichever is higher (Article 99).
High risk does not mean a ban. It is a set of duties for the provider and the deployer of the system: for example, under Article 26 the deployer assigns human oversight by people with the right competence and keeps the system’s logs for at least six months.
The Russian approach works differently: a framework law on large models, ethics and personal data. A comparison is handy to keep in view, and it is on the page “Legal regulation of AI in Russia”.
Justice in Annex III
Point 8(a) of Annex III: AI systems intended to be used by a judicial authority or on its behalf to help it research and interpret facts and the law and apply the law to a concrete set of facts. The same applies to similar use in alternative dispute resolution.
A court that buys such a system becomes its deployer and carries the duties of Article 26: use according to the instructions, human oversight and keeping logs. The developer of the system is liable as the provider.
The key test in the text of the point is who the system is intended for. Point 8(a) names a judicial authority and a dispute resolution body, so an assistant that a law firm’s lawyer uses to search case law or check a contract does not fall under it by the literal text. Check the position on a specific system against its intended purpose in the provider’s documentation.
Annex III has an exception under Article 6(3). A system is not considered high risk if it poses no significant risk to health, safety or fundamental rights and also:
1. performs a narrow procedural task;
2. improves the result of a human activity already completed;
3. detects deviations from earlier decision patterns and does not replace human assessment without human review;
4. performs a preparatory task for an assessment.
A system that profiles natural persons always stays high risk under Article 6(3). For courts, this means an assistant that prepares a draft decision falls under point 8(a), while narrow procedural functions may fall under the exception. The review of AI in courts around the world covers how courts in different countries already use AI.
New deadlines
The Digital Omnibus package moved the schedule. Duties for high-risk systems from Annex III, including justice, will apply from 2 December 2027 instead of 2 August 2026.
For systems from Annex I, meaning AI inside products with their own sector rules, the new date is 2 August 2028. The amending regulation came into force on 27 July 2026.
The delay covers only high-risk systems. The article on the EU AI law and Russia covers what in the regulation already applies, including the transparency rules.
For a lawyer, the delay gives time, not an exemption. During 2026–2027, a client that supplies AI solutions to the EU should define its role (provider or deployer), describe the purpose of the system and collect the documents for it.
Who in Russia it affects
The regulation also applies outside the EU. Under Article 2(1)(a), it covers providers that place an AI system on the EU market, wherever they are. Under Article 2(1)(c), it covers providers and deployers from third countries if the output of the system is used in the EU.
Three typical situations of Russian clients:
1. A company sells AI software to clients in the EU: it is a provider, and for a high-risk system it carries the provider’s duties.
2. A company uses AI to prepare materials that are used in the EU, for example in a dispute or in reporting to a European partner: check whether the system falls under Article 2(1)(c).
3. A company works only with clients in Russia: the regulation as a rule does not apply, but its requirements are a useful guide when choosing AI for legal work.
Under Article 26, the deployer of a high-risk system uses it according to the provider’s instructions, assigns human oversight by people with the right competence, training and authority, and keeps the logs the system records automatically for at least six months.
It makes sense to build these measures into the contract with an AI vendor now: an action log, roles and human oversight are needed even without the European regulation, and you will not have to rewrite the contract when you enter the EU.