Guide · AI and the law
How lawyers should use AI: 7 best practices to protect your case
Artificial intelligence is already in law offices: it drafts, summarizes files and finds case law in seconds. The serious question is no longer whether to use it, but how. Used well, it saves you hours; used badly, it has cost several lawyers a sanction. This guide gathers the best practices courts and bar associations now expect —including Colombia's Constitutional Court in ruling T-323 of 2024— so AI works in your favor.
The right question: not "do I use AI?", but "how do I use it?"
Banning AI in a law firm is as unrealistic as ignoring email. The edge no longer belongs to whoever uses it, but to whoever uses it with method. The difference between a lawyer who multiplies their productivity and one who ends up explaining to a judge why they cited a ruling that doesn't exist almost always comes down to these seven rules.
The 7 best practices
- Verify every citation against its primary source. The golden rule. No ruling, statute or citation generated by AI should reach a brief until you have confirmed, in the official source, that it exists and says what is claimed. You are responsible for the final product, not the tool.
- Never upload confidential client information to a general-purpose AI. What you paste into a public tool may be stored and used to train the model. Anonymize, minimize data, and keep sensitive information for tools with confidentiality and encryption guarantees. Privilege isn't suspended just because you used AI.
- AI assists; the judgment is yours. It's the same standard the Court set for judges in T-323: AI is complementary support, never a substitute for reasoning, evidence assessment or strategy. Delegate your professional judgment and you delegate your responsibility… but the sanction is still yours.
- Understand the tool before you trust it. There is a duty of competence: you must know what your system does well and badly (hallucinations, bias, training cutoff, whether it accesses real sources). Using AI you don't understand is flying blind.
- Prefer tools that show you the source. An AI that answers without saying where the information came from shifts all the verification risk onto you. Tools that link every statement to its original document let you check in seconds instead of hours.
- Be transparent and document your process. It's not about "citing the AI" as a source, but about being able to explain which part of the work was AI-assisted and how you verified it. That traceability protects you and respects the other party's right to contest.
- Know your jurisdiction's rules. In Colombia, T-323 of 2024 and the Judicial Council's guidelines already set the terrain. Rules evolve fast: stay current before adding a tool to your workflow.
The cost of getting it wrong isn't theoretical
AI hallucinations —false information with a flawless appearance— already cost cases and money. The case that woke up the profession was Mata v. Avianca (New York, 2023), where a court sanctioned two lawyers for submitting case law invented by ChatGPT. It wasn't isolated: in 2025, court reports documented dozens of briefs with fabricated citations across several countries, and Spain's High Court of Justice of the Canary Islands fined a lawyer for including up to 48 fake rulings suggested by AI in a single appeal. Citations with convincing docket numbers… of rulings that never existed.
The message for practitioners is direct: the problem isn't using AI, it's using it without verifying every citation against its real source.
What Colombia requires: ruling T-323 of 2024
Colombia's Constitutional Court set the country's first substantive position on AI in the justice system: the tool may assist, but never replace judgment, and it expressly warned about hallucinations, bias and lack of transparency. For the full case and what it means for your practice: Ruling T-323 of 2024: what every lawyer should know before using AI.
How HOFUSS solves it
At HOFUSS we build on that same logic: AI is a case-law research assistant, every citation is checked against its real source before it's shown, and the decision is always the lawyer's. If a reference can't be verified, we don't show it. That way rule #1 —verify every citation— stops being a manual chore and becomes part of the tool.
Sources
- Ruling T-323 of 2024 — Constitutional Court of Colombia (official text)
- ABA Formal Opinion 512 — Generative AI in legal practice (summary, UNC Law Library)
- ICAM best-practices guide for the use of AI in legal practice (Spain)
- Canary Islands High Court fines a lawyer for 48 fake AI-suggested rulings — CGPJ
- US court sanctions lawyers for fake case law created with ChatGPT — Diario Constitucional
This article is informational and does not constitute legal advice. Always verify official sources.