Case study: fixing an AI-generated shareholders’ agreement
Our corporate and commercial solicitors have seen a steady uptick in ‘Google law’ over the past few years, with a notable rise in recent months. This is the use of AI systems to draft complex legal documents.
While there is no question that AI is useful, there are several risks associated with using it to draft legal documents, as we recently discovered when a client came to us with an AI-generated shareholders’ agreement that was not fit for purpose.
Case in point: the risks of using AI to draft complex legal agreements
As artificial intelligence becomes more accessible, some clients are turning to AI tools to produce first drafts of legal documents, including shareholders’ agreements and heads of terms. At first glance, this can seem cost-effective: ask an AI programme to generate a long and complex agreement, then have a lawyer ‘lightly review‘ it.
However, as our Corporate and Commercial team recently discovered, this approach can create more work and more cost than expected.
Case study: using AI for a shareholders’ agreement
Our client invested in a company with an AI-generated shareholders’ agreement. At a glance, it was superficially impressive and contained some glossy jargon. The client wanted us to merely check over it to ensure it worked. However, when we looked closely, it contained inconsistent definitions, missing cross-references, conflicting clauses, outdated legal concepts, and commercially unworkable provisions.
One of our lawyers said the agreement was like a car which looked fine from the outside, but when you looked under the bonnet, it was missing the engine and a tank of fuel.
We found that the AI-generated agreement outlined the key points the parties wanted to include, but contained duplication and overlap, making it confusing. The parties wanted the documentation to be as simple and complete as possible. We therefore included the necessary language to enable the terms to be easily enforced and provide the parties with clarity on the details of what they were agreeing to.
Inaccuracies in the AI-generated agreement included:
- Using percentages that totalled over 100% (a simple error that a non-lawyer might pick up), and
- Not considering the risks of the interaction between the directors and shareholders in the decision-making of the company.
- This is an example of where AI is far behind in considering and mitigating the breadth of risks that might be relevant for an agreement, in light of all the specific facts and circumstances.
What are the risks of AI-generated legal documents?
Because AI models are trained on wide-ranging data rather than tailored to any jurisdiction, they can blend concepts from multiple legal systems or include clauses that do not reflect current law. Lawyers must then pick apart the document line by line, correcting errors, filling gaps, and restructuring the agreement so that it works in practice.
This can take significantly more time than drafting a tailored agreement from scratch, particularly where the AI has confidently included provisions that are subtly but critically wrong.
Crucially, using AI to draft an agreement will also mean that, if an issue arises down the track which was due to a fundamental error in the drafting, there will be no recourse for the client.
A law firm takes responsibility for the agreement they are reviewing and the consequences if it makes an error. AI does not.
What are the pros and cons of using AI for legal drafting?
AI can be a useful starting point, especially when considering your options or setting out key issues. However, there are several risks, especially with corporate and commercial legal agreements.
Pros
- Speed: AI can generate a rough structure quickly, particularly useful for agreements or contracts.
- Generating ideas: useful for outlining issues to consider or suggesting clauses for discussion.
- Cost-effective bullet points: helpful for early-stage thinking before instructing lawyers.
Cons
- Inaccuracy and inconsistency: AI can present legally incorrect or contradictory terms in a persuasive tone and draft a document with internal inconsistency.
- Hidden risks: no matter how much experience you have as a business owner or employer, errors may not be obvious to non-lawyers. These can leave you exposed to significant legal or commercial risk, particularly if you need to enforce your rights under an ineffective agreement.
- Potentially higher legal fees: reviewing and correcting an AI draft can take longer than producing a bespoke agreement. In the case of this AI-generated shareholders’ agreement, the client no doubt wished they had instructed a lawyer in the first place.
- Confabulation: we have seen AI rapidly and confidently produce its own incorrect answers to legal queries and only admit these are wrong when it is prompted by a qualified individual to reconsider with more detailed analysis.
What prompts should I use? Should my lawyer see them?
AI tools respond directly to the prompts they are given. If instructions are vague, incomplete or internally inconsistent, the output will be too.
For clients who choose to use AI as a starting point, one helpful step is to share the original prompts with their lawyers. These prompts often reveal the client’s priorities, concerns and assumptions, information that may not be obvious from the draft alone. Seeing the prompts can help lawyers quickly identify key issues and ensure those commercial points are properly captured in the final agreement.
Can I use AI to generate a legal agreement for my business?
AI can be a useful tool, but it has serious limitations. Currently, AI used for legal drafting cannot ensure compliance with current legislation, regulatory requirements or market practice. Nor can it undertake legal analysis or risk assessment.
As we have seen, it can give clients false security: they think that they have done the right thing in preparing an agreement, but may find it is unreliable if they try to enforce it.
However, looking forward, we can expect:
- Better integration with verified legal databases, reducing (but not eliminating) accuracy issues.
- Tools that can highlight uncertainties or areas requiring human review.
- More structured workflows combining AI efficiency with lawyer oversight.
But even as AI improves, it is unlikely to replace the need for specialist legal judgement, bespoke drafting, or an understanding of the commercial objectives behind each deal.
Final thoughts
AI can be a powerful tool and useful in specific circumstances. But it is not yet a substitute for legal expertise. In many cases, especially with complex documents such as shareholders’ agreements, using AI as a shortcut can end up being a false economy.
With proper guidance, AI can support the legal drafting process. However, the foundations of any binding agreement should still be laid by professionals who understand both the law and your commercial goals.
Contact our corporate and commercial lawyers
For expert legal advice for your business that aligns with your commercial goals and won’t leave you in a bind, get in touch. Contact our Corporate and Commercial team on 0117 325 2929 or fill out our online enquiry form.
