When is using AI a breach of contract?
Before using AI-enabled tools, you should check the contract and cross-reference that with your insurers. That was one of May Winfield’s warnings during her presentation at Digital Construction Week.

As a co-founder and chair of BIM4Legal and global director of commercial, legal and digital risks at Buro Happold, Winfield is an acknowledged expert on the interface between the law and digital processes and technology.
She warned the audience to check contracts, especially on public-facing AI tools: “Read the fine print. [The following fine print] is from a public-facing tool. On their website it states [that they have the right] to ‘copy, distribute, transmit, publicly display, publicly perform, reproduce, edit, translate and reformat’ captions, prompts, creations and other content provided. Basically, if you put in anything, they can do whatever they want with it forever and ever.”
She reinforced the need for a human-in-the-loop approach to AI use, and when artificial general intelligence arrives (as early as 2030 if pundits are to be believed), the need for an internal control framework. She noted that while artificial general intelligence may be as intelligent as a human, it has no ‘legal personality’ and thus cannot be sued.
Copyright and ownership continue to be an issue, Winfield said, both in terms of what is input into AI and the output generated. “If you input your designs, models or data into a public-facing tool, you are effectively throwing it into a public forum. You can’t delete it, you can’t remove it. Someone asking the right question can extract that data. If it’s your client data, if it’s your trade secret, that’s probably not a good idea.
“You don’t really know when you get output whether an element is in breach of copyright. It doesn’t matter if you take a design from an AI tool and change it or develop it, it’s [still] a reference point. It could be very embarrassing later if someone says, ‘that whole section of what you did belongs to me’: that could lead to legal liability.
“Most of your contracts, if not all, will say something like ‘I promise you, Mr Client, that everything I’m providing you either was created by me or I have a license to give it to you’. Now, if it was created by AI, can you hand on heart say that you have a copyright license to use it?”
Indemnity and confidentiality
Winfield noted that even closed, enterprise-level AI tools can cause breaches of confidentiality clauses. “You may say, ‘we’re not going to put confidential project or client data into a public-facing AI tool; we’ll just put it in the one that our company uses, and that’s fine, right?’ Not quite, because a lot of your contracts for confidential projects or certain clients say, ‘only the people working on this project can see the designs, can see the models, can see the data’; when you feed it into your AI, that means a colleague could extract it [and that’s a breach].”
She emphasised that professional indemnity insurance provides cover for the exercise of reasonable skill and care, but an insurer could consider failure to check AI outputs as reckless or negligent, invalidating the cover.
When using agentic AI, Winfield urged the audience to pressure-test the code behind the AI agents with as many scenarios as possible. “There may be a difference between the intention of a code and what happens in reality,” she warned, citing the Air Canada case where the airline’s chatbot went against company policy and effectively misinformed a customer about a fare.
Winfield highlighted another area where the use of AI goes against the law: GDPR. She said: “It’s worth bearing in mind that most AI tools are not GDPR-compliant. If you start feeding in personal data into your AI tools, speak to your GDPR officer, because you can’t delete [the personal data that you fed in], you can’t remove it, and that’s actually a breach of GDPR.”
Keep up to date with DC+: sign up for the midweek newsletter.