When AI gets the contract wrong, who’s actually responsible? The answer should change how you work

AI drafting professional responsibility is not as abstract as it is sometimes presented. When a lawyer uses AI to generate a contract and that contract contains an error — a wrong limitation of liability clause, a missing carve-out, an incorrect governing law — the question of who is responsible has a clear answer. It is the lawyer. The same applies whether the document was produced using a word processor, a template, or an AI drafting tool. Professional responsibility follows the qualified person who deployed the work, not the technology used to produce it.

Who is responsible? The answer from every professional regulatory body that has addressed the question is the same: the practitioner. This should not change whether firms use AI in document drafting — it should change how they use it.

AI drafting professional responsibility: what the regulatory framework says

The SRA’s guidance on AI makes clear that solicitors must maintain adequate supervision of AI-generated work and remain accountable for the advice and documents they issue. The ABA’s Model Rules similarly require competent, supervised use of technology in legal practice. Neither body has prohibited AI use in legal work — and nor should they. AI-assisted drafting, used properly, can improve consistency, reduce drafting time, and support more thorough review of complex precedents.

The practical implication is that a workflow designed purely around speed — AI generates, lawyer briefly reviews, document goes out — is not a compliant workflow if the review is not substantive. Speed is a benefit of AI drafting; it is not an excuse for reduced diligence.

Insight: It is worth considering whether your current AI-assisted document workflow is designed to support thorough review or to make review feel less necessary. These are very different things, with very different implications for professional liability.

The speed problem: why fast AI creates review pressure

One of the less discussed risks of AI-assisted drafting is the pressure it creates on review. If AI produces a first draft in minutes, there is an implicit expectation — from clients, from partners, from billing structures — that the overall turnaround should be faster. That pressure can compress the time available for meaningful review.

But the review is where AI drafting professional responsibility sits. The AI does not reduce the lawyer’s responsibility; it changes the nature of the task. Instead of drafting, the lawyer is reviewing and approving. That is still a substantive professional act, and it requires appropriate time and attention.

Firms that have not designed their AI workflows with this in mind risk creating a culture where speed is the measure of success, and review becomes a formality rather than a safeguard. That is precisely the scenario professional regulators are concerned about — and rightly so. For further context on how governance structures can help, see our piece on AI document governance for law firms.

What a properly supervised AI drafting workflow looks like

Proper oversight in an AI-assisted document workflow is not simply “a lawyer reads it before it goes out.” It is a structured process with defined checkpoints, clear accountability, and documented review. The following elements are characteristic of a well-designed workflow:

Approved source material: The AI should draw from a governed, up-to-date clause library rather than generating from general training data. This is fundamental to managing the quality of the output and limiting the scope of AI drafting professional responsibility exposure.

Structured review checkpoints: Different document types warrant different review requirements. A routine NDA reviewed by a paralegal may be appropriate in some firms; a bespoke financing agreement almost certainly warrants partner sign-off. The workflow should encode these distinctions explicitly.

Audit trails: The system should record who reviewed what, when, and what changes were made. This is both a risk management tool and a professional compliance requirement.

Escalation design: The workflow should make it easy to escalate when something falls outside the standard parameters. An AI that flags uncertainty is more useful than one that produces confident output regardless of complexity.

Insight: Firms often find that designing proper oversight into an AI drafting workflow surfaces questions they had not previously considered — about document ownership, review standards, and what “approved” actually means for a clause or template. These are valuable conversations to have before a problem arises rather than after. See also: how document automation integrates with your legal tech stack and managing legal precedents effectively.

Common mistakes to avoid

• Treating AI review as qualitatively different from template review. The professional responsibility for an AI-generated document is the same as for a template-based document. Firms that treat AI output as inherently more reliable, or as requiring less scrutiny, are taking on unnecessary risk.
• Optimising the workflow for speed rather than reliability. Speed is a genuine benefit of AI drafting, but it is a benefit that accrues to the firm. The client’s interest is in accurate, reliable documents. These are not in conflict — but only if the workflow is designed correctly.
• Using AI that draws from ungoverned source material. AI generating from general training data rather than a governed clause library introduces risk that is difficult to quantify and harder to defend.
• Not recording review activity. The absence of an audit trail does not change the professional responsibility — it just makes it harder to demonstrate that the obligation was discharged.
• Assuming AI errors will be obvious. AI can produce confident, well-formatted text that contains substantive errors. Review needs to be substantive, not just presentational.

XpressDox’s perspective: oversight is a feature, not a friction point

The firms that will use AI in document generation sustainably are those that treat oversight as part of the value proposition, not as a burden on top of it. A workflow that builds in proper review checkpoints, draws from a governed clause library, and maintains a full audit trail is not slower than one that does not — it is more defensible, more consistent, and more trusted by the lawyers who use it.

Speed without reliability is not an improvement. The goal is documents that are produced quickly and can be trusted. Building in AI drafting professional responsibility from the design stage — not as an afterthought — is what makes that possible.

Conclusion

The professional responsibility framework for AI-assisted document drafting is clear: lawyers remain responsible, regardless of the technology they use. The appropriate response is not to avoid AI — it is to deploy it within a workflow that is designed for proper oversight, governed source material, and documented review. Firms that get this right will benefit from both the efficiency of AI drafting and the professional standing that comes from demonstrable compliance.

If you would like to understand how document automation can be designed to support proper oversight rather than circumvent it, speak to the XpressDox team.

Frequently asked questions

Are lawyers legally liable for errors in AI-generated documents?

Yes. Professional regulatory bodies including the SRA and ABA have confirmed that lawyers retain full professional responsibility for documents they issue, regardless of whether those documents were produced using AI drafting tools.

What does adequate supervision of AI-generated documents mean in practice?

Adequate supervision means a qualified person reviews the output against the requirements of the matter, with sufficient time and context to identify errors. It is not satisfied by a cursory read of a well-formatted document.

Does using AI in document drafting increase professional risk?

It can, if the workflow is not properly designed. AI drafting professional responsibility is best managed through governed source material, structured review checkpoints, and documented audit trails. With these in place, AI-assisted drafting is no riskier than a well-managed template-based workflow.

Should firms have a specific AI policy covering document drafting?

Yes. A policy covering which AI tools are approved, what source material they draw from, what review requirements apply, and how decisions are documented is good practice and increasingly expected by regulators and clients alike.