Who Is Responsible When AI Makes a Mistake in Legal Work? Rules in the US, UK, Canada and Australia

Who is accountable when AI gets legal work wrong?

The lawyer is. No regulator in the US, UK, Canada or Australia has created a separate standard that shifts responsibility to the tool or the vendor. Existing duties of competence, confidentiality, supervision and candour apply to AI-assisted work exactly as they apply to work done by a junior lawyer or an outside provider.

What do the regulators actually say?

• United States: ABA Formal Opinion 512, issued in July 2024, applies the Model Rules on competence, confidentiality, communication, supervision and reasonable fees to generative AI.
• England and Wales: the SRA states that firms “will always remain accountable” where work is carried out through others, and cannot outsource reserved legal activities to unauthorised providers.
• Canada: the Law Society of Ontario’s April 2024 white paper says certain tasks require the licensee’s own judgment and cannot be delegated, that confidential information needs adequate safeguards, and that lawyers should bill only for time actually spent even when AI makes the work faster.
• Australia: the NSW Supreme Court’s Practice Note SC Gen 23, effective 3 February 2025, requires practitioners to understand the risks of generative AI, including fabricated citations and unverified material.

Why are lawyers cautious about AI?

Because the risk is theirs. The ABA’s 2024 Legal Technology Survey found 74.7% of lawyers named accuracy as their top AI concern, ahead of reliability (56.3%) and data privacy (47.2%). In Western Australia, the state Law Society’s 2025 survey found 72% of practitioners worried about confidentiality and privilege risks. Among US plaintiff firms, Supio found 99% would not use AI content they cannot verify.

Where does accountability break down inside a firm?

Mistakes rarely come from the tool alone. They come from gaps around it:

• a draft is treated as finished because it reads well
• nobody is named as the reviewer for AI-assisted work
• client data goes into a tool whose terms nobody checked
• the firm cannot show afterwards what the tool produced and what a person changed

Clio’s 2025 Legal Trends Report found 53% of legal professionals work somewhere with no AI policy or are unaware of one. That is where most of the risk sits.

How do firms build accountability into AI use?

1. Name the reviewer. Every AI-assisted output has one lawyer responsible for checking it before it leaves the firm.
2. Verify against the source. Citations, dates, medical facts and figures are checked against the underlying documents, not the AI’s summary.
3. Choose tools that show their sources. Output that links each statement to its source page is far quicker to verify.
4. Write a one-page policy. Approved tools, what data may go into them and when clients are told.
5. Keep a record. Save the prompt, the output and the reviewed version for anything that matters.

Do clients want to know when AI is used?

Increasingly, yes. Clio found 78% of clients want lawyers to disclose AI use, while 35% of legal professionals rarely or never do. A firm that can explain its review process in two sentences builds trust rather than raising doubt.

Where to start

Pick one AI-assisted task your firm already does and write down who reviews it, against what, and where the record is kept. If you cannot answer all three, close that gap first.

Sources

• Supio 2026 AI Adoption Gap report (press release)
• ABA Formal Opinion 512: Generative Artificial Intelligence Tools
• SRA: Lawtech providers Q&A
• Law Society of Ontario generative AI guidance, summarized by Osler
• NSW Supreme Court Practice Note SC Gen 23, reported by Australasian Lawyer
• ABA 2024 Artificial Intelligence TechReport
• Law Society of Western Australia 2025 Generative AI Survey
• Supio: plaintiff firm AI adoption
• Clio 2025 Legal Trends Report summary