# EU AI Act 2026: Risk Levels and Rules for Law Firms

> Learn how law firms can assess EU AI Act risk levels, apply transparency rules and plan compliance for chatbots, hiring tools and legal AI.

EU AI Act risk levels depend on each feature’s intended purpose. A law firm’s contract summarizer, hiring assistant and client chatbot can trigger different obligations. Check prohibited practices, high-risk uses, then transparency requirements; rules can overlap.

As of 3 October 2026, transparency requirements apply; main Annex III high-risk rules start on **2 December 2027** and Annex I product high-risk rules on **2 August 2028**. These reflect amendments enacted July 2026. See the Commission’s [AI Act overview](https://digital-strategy.ec.europa.eu/en/faqs/navigating-ai-act) and [current implementation timeline](https://ai-act-service-desk.ec.europa.eu/en/ai-act/timeline/timeline-implementation-eu-ai-act).

## Prohibited AI practices: identify uses that cannot proceed

Original prohibitions took effect 2 February 2025. They cover specified harmful manipulation, vulnerability exploitation, social scoring, untargeted facial-image scraping and certain biometric practices. The Act prohibits emotion recognition in workplaces and educational institutions except for medical or safety purposes.

Specific legal conditions, not product marketing labels, determine whether these categories apply. The Commission’s [explanation of prohibited practices](https://digital-strategy.ec.europa.eu/en/faqs/navigating-ai-act) provides the starting point.

Employers should flag proposals to infer employees’ emotions from facial biometric signals to measure engagement for prohibition review. Consent checkboxes create no additional exception.

The amended timeline prohibits systems generating non-consensual sexual deepfakes or child sexual abuse material from **2 December 2026**. [Official timeline](https://ai-act-service-desk.ec.europa.eu/en/ai-act/timeline/timeline-implementation-eu-ai-act).

## High-risk AI: classify the purpose and its effect

High-risk classification has two main routes. One concerns products or their safety components under Annex I legislation and subject to qualifying third-party conformity assessment. AI in medical devices can qualify; ordinary convenience features do not automatically qualify. The other covers specified Annex III uses. [Article 6](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-6).

A law firm’s assistant ranking paralegal applicants is a recruitment use requiring Annex III assessment. Helping judicial authorities research facts and apply law to cases is also a listed purpose, including comparable use in alternative dispute resolution. Ordinary legal drafting for private clients is not automatically high-risk. [Annex III](https://ai-act-service-desk.ec.europa.eu/en/ai-act/annex-3).

Some Annex III tools, including narrow procedural or preparatory tools, may qualify for Article 6(3)’s exception if they meet its conditions and pose no significant risks. Human involvement alone does not suffice. Annex III systems profiling individuals remain high-risk; providers claiming exceptions must document their assessments. [Classification requirements](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-6).

Request vendors’ intended-purpose statements and classification rationales. Record decisions outputs influence; reject “administrative assistant” as a legal classification.

## AI transparency obligations: chatbots, deepfakes and published text

Article 50 transparency rules took effect **2 August 2026**, per the Commission’s [enforcement announcement](https://digital-strategy.ec.europa.eu/en/news/commission-starts-enforcing-ai-act-rules-and-new-transparency-requirements-2-august).

Providers must design AI client-intake chatbots to disclose AI interaction to users unless context makes it obvious. A clear notice at first interaction is practical. This duty alone does not make chatbots high-risk.

Providers of synthetic text, audio, image or video generators generally must mark outputs in a machine-readable form and make them detectable as artificial, with Article 50(2)’s exceptions, including certain standard editing functions. Deployers must disclose deepfakes, subject to the article’s exceptions and special provisions.

AI-generated text published to inform the public on matters of public interest requires disclosure unless exempt. The editorial exception requires human review or editorial control **and** an editorially responsible person or entity. Firms publishing AI-assisted legal updates should document that responsibility. [Article 50](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-50).

Certain providers of synthetic-content systems marketed before 2 August 2026 must comply with Article 50(2) by **2 December 2026**. This does not postpone all transparency obligations. [Transition timeline](https://ai-act-service-desk.ec.europa.eu/en/ai-act/timeline/timeline-implementation-eu-ai-act).

## AI providers and deployers: roles and general-purpose AI duties

Firms using purchased software are ordinarily deployers. Developing or commissioning a system and putting it into service under the firm’s name can confer provider status, including for internal use. [Article 3 definitions](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-3). Repurposing a system for high-risk use can create provider responsibilities. [Article 25](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-25).

When high-risk requirements apply, providers must manage conformity assessment, technical documentation, risk and compliance throughout the system’s lifecycle. [Commission guidance](https://digital-strategy.ec.europa.eu/en/faqs/navigating-ai-act). Deployers must follow instructions, assign competent people for oversight, monitor operations and retain automatic logs they control for at least six months unless law provides otherwise. [Article 26](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-26). Build these requirements into procurement before they take effect.

Separate duties for general-purpose AI (GPAI) models cover technical information, copyright policies and training-content summaries, plus risk-management duties for systemic-risk models and some open-source exceptions. Chatbot subscriptions alone do not make firms underlying model providers. [GPAI obligations explained](https://digital-strategy.ec.europa.eu/en/faqs/general-purpose-ai-models-ai-act-questions-answers).

GPAI duties began 2 August 2025; providers of models marketed earlier have a transition until 2 August 2027. [GPAI guidance](https://digital-strategy.ec.europa.eu/en/policies/guidelines-gpai-providers).

## Lower-risk AI: retain controls and support staff AI literacy

Extracting renewal dates for paralegal verification may fall outside prohibited and high-risk uses. Lower-risk or minimal-risk use can offer a manageable start, but classification leaves confidentiality, data protection and accuracy concerns unresolved. Check access controls, retention terms and source-document verification. Legavima’s [CLM software evaluation guide](https://legavima.com/blog/clm-software-evaluation-checklist-for-legal-and-procurement/) offers related procurement checks.

Under amended Article 4, providers and deployers must support staff AI literacy development, considering their knowledge, experience and use context. They need not guarantee individual literacy levels. July 2026 amendments revised this duty without abolishing it. [Commission AI literacy Q&A](https://digital-strategy.ec.europa.eu/en/faqs/ai-literacy-questions-answers).

Train legal teams on fabricated citations, confidential information and escalation when outputs conflict with source documents. Record your guidance and revisit it when features or responsibilities change.

## Next steps for law firms

Register each feature’s intended purpose, affected people, provider, internal owner and deadline. Attach the vendor’s classification explanation and record each feature’s actual organizational use.

Ask qualified counsel: Is this use prohibited or within Annex III? Are claimed exceptions documented? Have customization or repurposing shifted provider responsibilities? Which Article 50 notices or markings apply? Does a transition cover this system? What additional privacy, employment or professional duties apply?

Assign actions and review dates accordingly. Reassess classification when document assistance shifts to influencing decisions about people.

---

[View the canonical page](https://legavima.com/blog/eu-ai-act-risk-levels-in-2026-which-rules-apply-to-your-ai-f/) · [Browse llms.txt](https://legavima.com/llms.txt)
