AI legal advice is increasingly integrated in our daily life, at home and in business. It brings efficiencies and cost-savings and we’ve seen a significant uplift in the use of AI legal advice in place of professional advisers. AI is increasingly used to draft contracts, review documents and provide legal guidance.
While AI can be a great user-friendly starting point, reliance on AI generated material carries risk. We are seeing clients requesting a review of ‘DIY’ legal documents produced with AI and almost always, rectifying the issues created is more time-consuming and costly than it would have been if we had been involved earlier. Below are some key risks to be aware of when using AI tools for legal or commercial purposes.
1. Hallucinations and inaccuracies
AI models predict based on training data and probable outcomes; models anticipate the text that should follow the prompt based on previous information. Inevitably this sometimes goes awry producing hallucinations, where the model generates false but very plausible information that is easy to miss.
Common issues with AI-generated legal documents
AI-generated documents can:
- rely on outdated law or principles
- result in documents missing crucial clauses
- cite false court citations and legislative references
- be tailored to the wrong jurisdiction
- fail to reflect the commercial agreement between the parties
- contain inconsistencies or conflicting clauses.
We’ve seen investment documents drafted back to front, where the minority shareholder has been given majority shareholder style voting and decision-making. AI can produce convincing looking documents but when closely reviewed, they may not do what they need to and users can be left exposed.
2. Confidentiality
Businesses should be careful about inputting confidential or sensitive information into public AI systems. Information shared with public AI tools can be stored and used to train the AI model, which often does not have appropriate protections in place. Inputted information can be shared with human reviewers to verify accuracy and once stored, can be difficult to delete.
Legal privilege
Recent case law raises questions over legal privilege (the confidentiality of lawyer-client communications) and information shared on public AI models. In the recent UK v Secretary of State for the Home Department [2026] UKUT 81 (IAC), it was suggested that once privileged legal information is shared with public AI e.g. ChatGPT, this constitutes a waiver of legal privilege and the information is no longer confidential.
Data protection
In addition to leakage of confidential information, sharing sensitive personal data could breach the General Data Protection Regulation (GDPR) and many AI models lack the necessary data processing agreements to effectively safeguard this data.
3. Ownership
Copyright and AI-generated content
AI legal advice also raises important questions around ownership. Ownership of material created by AI remains a topic of debate as AI relies on learning from existing works. In the US, works created by AI are not eligible for copyright. In the UK, the author is deemed to be the person that made the necessary arrangements for the creation of the work. This remains an evolving area of law and records of input prompts should be kept to help preserve ownership claims.
Infringement risk
As public AI learns from inputted (and possibly copyrighted) material, using it to draft legal documents or contracts may result in something similar enough to existing protected material that it will create infringement issues.
4. Lack of human judgement
Generic output and commercial blind spots
AI produced documents are generic. They are not able to replicate company values or strategy and sometimes miss the commercial context. AI models may fail to include business specific provisions or apply all sector-specific legislative obligations.
Bias and transparency
As AI models are trained on historical data, this can reflect past discrimination and generated output can be inherently biased. This can create issues in drafting policy documents, producing prejudiced results. Due to the nature of many AI models operating as ‘black boxes’ it is very difficult to review the reasoning behind AI-generated documents or advice.
5. Liability
Who is responsible?
Liability for breach of confidential information, IP infringement or contradictory contractual provisions generally remains with the AI user. Insurance may not cover AI errors.
Insurance gaps
Insurers are aware of the issues that reliance on AI can create and may offer specific AI-related cover or carve it out from general cover. Businesses need to be aware of these potential gaps and ensure coverage is adequate.
AI has been a gamechanger across many professions and the law is no exception. It is ultimately software and should be used with discretion, rather than relied on in place of trained and experienced advisers. The risks of inaccurate drafting and incorrect advice significantly outweigh potential short-term savings. We’d always recommend using experienced professionals to protect yourself and ensure your documents do exactly what you need them to.
If you are concerned about AI legal advice or a document drafted using AI, you’re not alone. Our corporate and commercial team works with clients every day to review, advise and put things right. Get in touch and we’ll talk it through. Contact us today.
We publish blogs and social media posts to give a general overview of legal and commercial issues, relevant at the time of publication, which we hope you will find interesting. Please note that legal rules often change depending on the specific facts of a situation. The law also changes over time following changes in legislation or new court cases. We do not actively update our blogs or posts once they are published to reflect changes in the law.
As such, our blogs and posts are not intended to advise you on the law and must not be relied upon as legal advice. If you require advice on a particular issue then please contact us and we will be pleased to help.