AI as an External Legal Department
review by an ARROWS attorney, backed by our full liability and professional indemnity insurance
AI as an external legal department means that lawyers from ARROWS connect artificial intelligence to the client's systems, but they personally review every output before sending it and remain liable for it. It is suitable for companies across various industries, but not automatically for every one—this is preceded by an initial analysis of the agenda and security requirements. In this article, you will find out how the connection works, who is liable for any errors, and for whom it makes sense.
How to Replace Lawyers with Artificial Intelligence?
What "AI as an External Legal Department" Means in Practice
Companies across various sectors—from manufacturing and trade to services, IT, and construction—typically handle a recurring set of similar tasks: contracts with suppliers and customers, claims, general terms and conditions, employment law queries, GDPR, and inspections. They either have a full-time in-house lawyer for this or send the agenda to a law firm on a case-by-case basis, waiting days for a response.
The "AI as an External Legal Department" model stands in between. For each recurring type of client agenda, the Prague-based lawyers at ARROWS first create a playbook—a fixed procedure for assessing and handling the matter—and the AI prepares the initial draft based on it. A lawyer always reviews and supplements the output before the system sends it. Moreover, ARROWS' solutions are MCP-first, meaning they work with the latest approaches and security protocols. The client thus gets speed comparable to an in-house team combined with the expertise and liability of a law firm.
This is not a one-off experiment. The lawyers at ARROWS first used AI tools for several years on their own internal agenda and only began offering them to clients based on this experience—managing partner Jakub Dohnal summarized his first-hand experience in the article Lawyer vs. Algorithm. For some time now, they have also been operating the Cukuletu system for developers, which was covered by Hospodářské noviny.
This is not the same as a generic "AI lawyer" or a chatbot that offers ready-made legal solutions on its own. The Czech Bar Association, in its opinion of September 9, 2026, explicitly confirmed that such an independently acting tool would materially meet the definition of a legal service, which may only be provided by a lawyer. It is precisely the lawyer's review of every output that distinguishes the AI-supported external legal department of ARROWS from similarly-named tools on the market.
A Startup Won't Cover Your Damages, a Law Firm Must
Using AI does not transfer responsibility for the content of the legal service to the technology provider. A lawyer from ARROWS is always the person providing the legal service and bears full responsibility for it—just as if they had written the document themselves from scratch.
AI does not replace the decision-making and strategic role of a head of legal or a lawyer—this involves judgment, negotiation experience, and responsibility that cannot be delegated to software. The goal of the model is to strengthen the in-house team, not to bypass or replace it: AI eliminates routine tasks, while decision-making remains with people. Unlike a person, however, AI doesn't take vacations, get sick, or resign, and thus handles the routine part of the agenda without downtime.
Who bears the risk is also crucial: if a company sets up AI tools itself, the responsibility for any errors remains with it. In the external legal department model, however, the liability is borne by ARROWS and its insurance coverage of up to CZK 400,000,000—not the CFO or head of legal who approved the AI implementation in the company. Managing partner Jakub Dohnal also discussed what makes a lawyer irreplaceable and where AI excels in the Legal Code podcast: Where a Lawyer is Better Than Artificial Intelligence.
This also applies to confidentiality. The lawyer's duty of confidentiality under Section 21 of the Act on Advocacy and the protection of confidential information under Section 3a of the same Act are not altered by the use of AI in preparing documents. The lawyers at ARROWS therefore only work with tools where they have a contractual guarantee that client input data is not used for training models and is protected at a level appropriate to the sensitivity of the matter.
Frequently Asked Questions about Liability for AI Outputs
Who is liable if an error appears in an AI-generated document? The law firm ARROWS is liable for the output as a whole. A specific lawyer always reviews and modifies it before sending it to the client, relying on proprietary data, verified procedures, and ARROWS' internal directive for working with AI. Furthermore, this liability is covered by ARROWS' professional indemnity insurance up to CZK 400,000,000.
Can a lawyer overlook an error? The risk of human error can never be completely eliminated, even without AI. That is why ARROWS has an internal directive that precisely specifies when and how AI output is verified and what data may be entered into which tool.
Does the Czech Bar Association approve of this model? Yes. The current opinion of the Czech Bar Association (ČAK) from September 9, 2026, explicitly permits the use of AI by lawyers, provided that the lawyer maintains independence of judgment, confidentiality, and responsibility for the output—which is precisely the principle on which ARROWS builds its model.
How Integration with Your Systems Works and Where It Makes Sense to Deploy Agents
The integration doesn't start with a blanket connection of everything to everything. The implementation proceeds in three steps:
Initial Analysis – The lawyers at ARROWS will review the client's regular agenda with them to identify specific areas where AI can genuinely save time.
Pilot on One Type of Agenda – ARROWS will design a playbook for it and test it in operation; the rest of the agenda continues to be handled in the traditional way for now.
Expansion Based on Pilot Results – Only after successful verification in the pilot phase will AI agents and integration with the client's systems be extended to other agenda items based on additional playbooks.
Thus, handling the routine part of the agenda never begins by connecting all systems at once, but with a small, easily evaluable step. It's also true that this is not just about reducing costs for the legal team, but also for administration.
The most challenging part of the integration is usually not filling out forms or working with templates—many tools on the market can handle that in a basic form. The real work lies in determining what data is allowed to flow between systems, how it is secured during transfer, and who has access to it throughout the entire processing period.
Communication then runs through the channel the client already uses—email, Teams, Slack, or another corporate tool—and through a client portal where the status of each request is visible from submission through AI processing to review and dispatch by a lawyer. ARROWS adapts to how the client communicates, not the other way around.
For a company considering whether to hire a full-time in-house lawyer or entrust its agenda to a law firm in this manner, it means in practice that they pay for expertise and capacity based on the work actually performed, not for a full-time position. The routine part of the agenda is accelerated and made cheaper by AI, while more complex matters are handled by the same legal team that has long-term familiarity with the client's agenda.
AI-Replaced External Legal Department and Potential Risks. | How ARROWS Law Firm Solves It (consultation@arws.cz) |
|---|---|
Uncontrolled data transfer between client systems and the AI tool | We transfer data only through a verified interface with contractually excluded model training and a clearly defined scope of access |
Erroneous or factually incorrect AI output (so-called hallucinations) without review | Every output is verified and edited by a specific lawyer from ARROWS before being sent |
Unclear who bears responsibility when something goes wrong | The legal service is always provided by a named lawyer from ARROWS, whose liability is covered by insurance up to CZK 400,000,000 |
Unnecessarily slow handling of routine agenda (contracts, queries, claims) | Integration with your systems and a shared client portal reduce turnaround time from days to hours |
Uncertainty whether AI deployment complies with the rules of the Czech Bar Association | We proceed according to ARROWS' own internal directive, which follows the current opinion of the Czech Bar Association (ČAK) of September 9, 2026 |
Industry-specific security requirements (ISO standards, certifications, security clearances) that a standard AI deployment does not cover | The pre-deployment initial analysis will verify which certifications and standards your industry requires, and we will adapt the setup accordingly |
Who the Model is For and What the Initial Analysis Includes
We do not automatically deploy the AI-supported external legal department for every client. An initial analysis is conducted first, in which lawyers from ARROWS map the type and sensitivity of the agenda, the volume of recurring tasks, and which systems and data should be connected to the AI tools.
The analysis also includes the specific security requirements of the particular industry. A company in the financial sector, healthcare, public procurement, or the defense and security industry usually has different obligations regarding data and system security than a company in manufacturing, trade, or general services—typically involving its own security certifications, clearances, or standards like ISO/IEC 27001 for information security management. The lawyers at ARROWS will adapt the AI tool setup and integration with the client's systems accordingly, or recommend that a specific part of the agenda not be processed via AI at all. Furthermore, connecting AI to corporate systems may create its own obligations for the client under the European AI Act.
Managing partner Jakub Dohnal has long been focused on the regulation of artificial intelligence—he commented on the upcoming AI Act for Forbes back in 2023 (Europe will keep an eye on artificial intelligence. Will the world's first regulation be created here?) and then summarized the current obligations in the article AI in companies under EU supervision: new obligations 2026.
Development and security are not one-off matters. In 2023, ARROWS was the first law firm in the Czech Republic to establish its own specialized AI team; today, it consists of four developers and two UX testers who have been developing and testing AI tools for four years, and external security engineers who regularly assess the security of the solutions.
The work of this team was also recognized in the Law Firm of the Year competition, where ARROWS in 2024 received the Innovative Firm of the Year award. The assessment by external security engineers is part of the initial analysis for clients with higher security requirements.
The initial analysis also serves to ensure the model is not deployed where it doesn't make sense. It is typically not a suitable solution if:
the company does not have a recurring agenda and each case is unique in its own way—AI will not save time or costs in such a scenario;
the company wants to completely replace its legal department, not strengthen it—a lawyer always remains part of the process for every output;
the company is prohibited from connecting its data to any external tool for regulatory or security reasons—in such cases, the lawyers at ARROWS will recommend a different method of cooperation.
Only based on this analysis is a decision made on where to deploy AI agents, where to stick with the traditional process, and what contractual and technical guarantees the integration will require.
Final Summary
AI as an external legal department gives companies across industries speed comparable to an in-house team, without them having to deal with recruitment, substitutes, or the expertise of a single person for everything. However, it is not an automatic solution for everyone—it is always preceded by an initial analysis of the agenda, data, and security requirements of the specific industry. The key is that AI remains only a tool for processing documents—the legal service is always provided by, and its content is the responsibility of, a specific lawyer, who is also protected by the duty of confidentiality and ARROWS' insurance coverage of up to CZK 400,000,000.
If you are considering where an AI-supported external legal department could genuinely save you costs on an in-house team, we will discuss your agenda with you, review security and industry requirements, and propose where to deploy AI agents and integrate them with your systems in a way that makes sense. Contact ARROWS at consultation@arws.cz.
What You'll Want to Know About Replacing Lawyers with Artificial Intelligence
What is MCP, which ARROWS mentions in connection with integrating AI with client systems? MCP (Model Context Protocol) is an open standard for securely connecting AI tools with corporate systems and data. The lawyers at ARROWS use it where it makes sense to connect AI directly to a specific client system. Will AI replace the classic corporate lawyer? No. AI speeds up the processing of routine agenda, but the expert and strategic work, as well as the review of outputs, is always done by a lawyer from ARROWS. We would be happy to show you the difference on your specific agenda.
For what size company and what industry is this model suitable? It makes the most sense for companies with a recurring agenda across industries that do not want a full-time in-house legal department but need faster responses than with the classic case-by-case approach. For regulated industries (financial sector, healthcare, public procurement, defense and security industry), it also depends on which ISO standards and security certifications the company must adhere to—this will be verified by the initial analysis.
What happens to the client's data after the cooperation ends? The handling of data is governed by the legal services agreement and contracts with the AI tool providers; we will set the scope of data retention and deletion according to your requirements.
How much does such a solution cost compared to an in-house lawyer? It depends on the scope and type of agenda you want to handle through the AI-supported external legal department. We will prepare a specific proposal and cost comparison for you.
How to get started? We will review your current agenda with you, show you where to deploy AI agents and integrate them with your systems, and where, on the other hand, to stick with the traditional process.