Will AI become the new lawyer? When ChatGPT will take away lawyers' jobs and what IT companies should do

Why "AI Lawyer" is not tomorrow

Generative models work differently than a human lawyer: they do not “know” the norm, but predict the next words. Therefore, where the relevance of the norm, jurisdictional accuracy and context are required, they are easily mistaken. A prime example is tax changes: there are still answers where “military duty is 1.5%”. In fact, from December 1, 2024, the general rate is 5% (with some exceptions), and this is confirmed by official explanations from the tax service. If the AI ​​​​is not updated or does not have access to sources, you will receive outdated advice that costs money.

Another systemic flaw is “hallucinations”: fictional facts and precedents that sound plausible. Lawyers have already been fined for this: in Mata v. Avianca, a New York court imposed sanctions for fake links to “cases” generated by a chatbot; similar cases continue to multiply in different jurisdictions, and courts explicitly warn against the uncritical use of AI.

Therefore, the answer to the provocative question “when will AI take away lawyers’ jobs?” is this: AI will not take away, but will redistribute them. Purely mechanical tasks without responsibility will disappear — those who manage AI as a tool, take on legal risk assessment, and bear professional responsibility will survive.

Where ChatGPT already adds value

In legal practice, AI is consistently useful as a “first drafter” and assistant when we:

• We generate templates and wording options for typical documents (NDAs, security policies, job offers, standard annexes to contracts). Next, a strict legal redline and localization to Ukrainian law/English law if necessary.
  
  
• We do "issue spotting": a quick list of risks in an incoming contract with a client/contractor so that the lawyer can immediately see areas of attention (IP rights, SLA, fines, confidentiality, personal data, sanctions, export control).
  
  
• We summarize large amounts of text: tender documentation, internal policies, correspondence — to speed up the entry into context.
  
  
• We create checklists and playbooks: “due diligence questions”, “package of documents for team relocation”, “Data Subject Request procedure”.
• We analyze options for negotiating positions: we offer neutral compromise formulations to save time on agreement.

The key point: in all of these scenarios, the last word is with the lawyer, not the model.

Where ChatGPT is dangerous without safeguards

1. Relevance of norms. Ukrainian tax and currency regimes change frequently. The answer “1.5% military levy” of the 2023 model in 2025 is already incorrect and potentially unprofitable.
   
   
2. Jurisdictional substitutions. The model easily mixes US/EU/UK/Ukraine approaches (e.g., regarding at-will employment, consideration, or implied warranties), which makes the advice unusable.
   
   
3. Fictional references (hallucinations). This is not "case loading", but a reputational and procedural risk - from rejection of the submission to sanctions.
   
   
4. Privacy and privilege. Sending customer data to a public model is a bad idea. Private instances, anonymization, and contractual guarantees are needed.
   
   
5. Lack of legal qualifications. The model does not bear professional responsibility, does not evaluate evidence, and does not choose a procedural strategy.

How lawyers are implementing AI into the legal workflow

Lawyers approach gradually and with technical and procedural safeguards:

• Separate environment: corporate version with restrictions, logs, and no training on our data.
  
  
• RAG approach: the model responds only based on their library of templates, policies, current regulations, and internal comments; references to sources are required.
  
  
• Red zones (no-go): no "clean" advice on taxes, employment, sanctions, or litigation strategy without a partner review.
  
  
• Rules for prompts: no PII/trade secrets without anonymization; required context (jurisdiction, counterparty type, scope).
  
  
• The “two-eyes” procedure: each AI draft undergoes a human redline; in complex cases, “four eyes.”
  
  
• Fact verification: any quotes or norms - through official databases/primary sources; no "court decisions" with AI without checking in the registers.
• Client training: short guidelines for client teams so that their lawyers/operators can also use the tools without risks.

What IT companies should do today

1. Adopt internal AI policies. Define what data can and can’t be entered; who approves use; how requests are logged.
   
   
2. Update contract templates. Add provisions on the use of AI by contractors (privacy, IP, liability for model errors, prohibition on training on your data).
   
   
3. Implement a “legal intake bot.” Let AI gather facts and generate a resume for a lawyer — this shortens the response cycle without compromising quality.
   
   
4. Validate critical regulations. Taxes, employment, personal data (GDPR/Ukrzakon), sanctions — only with verification in primary sources or through an external legal team.
   
   
5. Learn from others’ mistakes. Court cases involving fake links are not “American tales,” but a warning to every jurisdiction. Courts around the world are already publicly demanding responsible use of AI.

Bottom line: there will be no “replacement”, there will be synergy

ChatGPT will not become a “new lawyer,” but will become a new working tool for a lawyer — as familiar as a document editor or a contract management system. Lawyers who ignore AI gradually lose in speed and cost. Lawyers who delegate decisions to AI without verification lose in reliability and are held accountable with money and reputation. The third way wins: a person sets a task, AI accelerates it, a person checks and is responsible.

And for IT businesses, the main message is simple: implement — but with the rules of the game. And remember: if ChatGPT confidently says today that the military fee is 1.5%, tomorrow it will just as confidently "find" you non-existent case law. Your processes should ensure that none of these mistakes become your risk.