Tuesday evening, partners’ meeting. The firm has been using AI for contract review and research in its own files for six months. Then a partner asks: “Under the AI Act, are we actually high-risk deployers? And who here is responsible for that?” Silence. Everyone has heard of the AI Act, hardly anyone knows what it means in concrete terms for their own firm.
The good news first: for most law firms and legal departments, the AI Act is manageable. But it requires clarity on three questions: which AI do we use, in which role and for what purpose?
- Law firms and legal departments are as a rule deployers, not providers of AI systems.
- Typical legal AI (research in your own files, contract review, drafting) is as a rule not a high-risk system.
- Relevant today: ruling out prohibited practices, AI literacy in the team, transparency obligations under Art. 50 (since 2 August 2026).
- High-risk obligations under Annex III apply under the Digital Omnibus only from 2 December 2027. They affect law firms more through areas such as recruitment than through legal advice.
- The AI Act replaces neither § 203 of the German Criminal Code nor the GDPR. For professionals bound by secrecy, these often remain the stricter benchmark.
Step 1: Clarify your role, provider or deployer?
The AI Act assigns obligations by role. The provider is whoever develops an AI system and places it on the market under their own name. The deployer is whoever uses it professionally under their own responsibility.
You are the deployer
You use a ready-made platform. The provider obligations lie with the vendor.
You can become the provider
You build an internal assistant on a language model or your own tool for client requests in a protected area of your website. Then you can become the provider yourself, with correspondingly more obligations.
Large firms and legal departments with their own legal tech unit in particular should draw this line consciously. A “small internal project” can shift the role.
Step 2: Determine the risk class. Is legal AI high-risk?
The worry that legal AI is automatically high-risk is widespread but usually unfounded. For the legal sector, Annex III of the AI Act mainly lists AI systems used by judicial authorities or on their behalf to research, interpret and apply facts and the law, as well as comparable uses in alternative dispute resolution.
High-risk possible
- AI prepares decisions as an arbitral tribunal or in mediation
- AI pre-screens applications for associates (Annex III, employment)
- The legal department uses AI in recruitment for the whole company
Typically not high-risk
- Research in your own files
- Contract review
- Drafting a memo in the firm
For high-risk systems under Annex III, the obligations under the Digital Omnibus (Regulation (EU) 2026/1744) only apply from 2 December 2027. That is time for a proper inventory, but no reason to put it off.
Step 3: Meet what already applies today
Regardless of high-risk status, obligations already apply today that affect every law firm:
| Obligation | Since | What it means for law firms |
|---|---|---|
| Prohibited practices (Art. 5) | February 2025 | Hardly relevant in practice, but rule them out consciously once in the AI inventory. Fines of up to €35 million or 7% of turnover. |
| AI literacy (Art. 4, as amended by the Digital Omnibus) | February 2025 | Staff who use AI must know what it does, where its limits are and how to check results. |
| Transparency (Art. 50) | 2 August 2026 | Tools for clients, for example in a protected area of the website, must disclose that an AI is answering. AI texts published on matters of public interest must be disclosed, unless a human has reviewed them and taken editorial responsibility. |
For client work itself, AI literacy is what matters most. It is more than a compulsory training session. It is the basis for lawyers checking AI results the way their professional duty of care requires anyway.
Step 4: Think AI Act, professional law and GDPR together
For professionals bound by secrecy, the AI Act is only one of three levels, and often not the strictest:
§ 203 StGB and § 43e BRAO
Govern the conditions under which client data may be passed to service providers.
GDPR
Requires a legal basis, a data processing agreement and additional safeguards for third-country transfers.
AI Act
Adds requirements on literacy, transparency and human oversight.
An AI tool that complies with the AI Act but transfers client data to the US or uses it for model training is still not an option for a law firm.
Step 5: Anchor human oversight in everyday work
The AI Act relies on human oversight. In everyday law firm practice, this means: no AI result goes to clients, to court or into a contract unchecked. But this only works if checking is practical.
This is where the tool’s architecture decides. An AI that only delivers plausible-sounding answers makes oversight expensive, because every statement has to be researched again. An AI that backs every statement with the specific passage makes it practical: you check exactly the passages that matter.
Checklist: seven questions for your firm or legal department
- InventoryDo we have an inventory of all AI tools used here, including private accounts?
- RoleFor each system, are we the deployer or the provider?
- High-riskDo we use AI in a high-risk area, for example recruitment?
- PolicyIs there an AI policy that sets out which tools are approved for which data?
- TrainingAre staff trained to check AI results critically?
- VendorDoes the vendor meet the requirements of § 203 StGB, § 43e BRAO and the GDPR?
- ResponsibilityIs it clear who in the firm is responsible for AI compliance?
Where aiuno comes in
aiuno is built for law firms, legal departments and regulated industries. For the requirements of the AI Act, professional law and the GDPR, this means:
| Requirement | How aiuno handles it |
|---|---|
| Clear roles | aiuno is the provider of the platform. You use it as the deployer and as a rule do not become a provider yourself. |
| Human oversight that works | Every answer is backed by document, section and passage. One click takes you to the original. |
| Confidentiality by design | Development and hosting in Germany, no transfer to the US, no training on your content, zero data retention and a confidentiality obligation under § 203 StGB. |
| Control over the infrastructure | Dedicated infrastructure or operation in your own data centre on request. |
| Working with your own files | Research, contract review with the Legal Crew and workflows based on your firm’s own knowledge. |
Conclusion
The AI Act does not turn legal AI into a high-risk project. But it requires what good law firms should deliver anyway: an overview of the tools in use, trained teams, verifiable results and clear responsibility. Anyone who chooses a tool that has confidentiality and source references built in from the ground up has already done most of the work.
And the partner from the Tuesday meeting? Her firm is a deployer, not high-risk, and above all needs one thing: a clearly named person who keeps the overview.
- Regulation (EU) 2024/1689 (AI Act), in particular Art. 4, 5, 25, 26, 50 and Annex III
- Regulation (EU) 2026/1744 (Digital Omnibus)
- § 203 StGB, § 43e BRAO
As of November 2026. This article is for general information and does not replace legal advice in individual cases.
Frequently asked questions
Is ChatGPT and the like banned in law firms under the AI Act?
No. The AI Act does not ban general AI assistants. Whether using them with client data is permissible depends mainly on § 203 StGB, § 43e BRAO and the GDPR.
Do we have to inform clients when we use AI?
The AI Act does not generally require this for internal use in client work. A notice is mandatory when clients interact directly with an AI, for example through a tool for client requests in a protected area of the website. Transparency towards clients under professional law and contracts has to be assessed separately.
When do the high-risk obligations apply?
For systems under Annex III, from 2 December 2027 under the Digital Omnibus.
Who is responsible in the firm?
The AI Act does not prescribe a specific function. A clearly named responsibility makes sense, for example with a partner, knowledge management or IT, drawing on data protection and professional law expertise.