Your creative team finished the pitch deck last Tuesday. An AI tool wrote the first draft of the headline copy. The designer refined it, the strategist added the data slides, and the deck went to the client as the agency’s work. Nobody thought to ask whether that arrangement needs to be in the contract. Nobody has told the client either.
It might not matter this time. But if that client later disputes the deliverable, or asks directly whether AI was involved, you will be managing two separate problems at once. Who owns the output, and why did nobody say anything. Both are fixable before they happen. Neither is easy to fix after.
Two questions decide whether AI becomes an agency asset or a liability. Both belong in the contract and the client conversation, not in the tool stack.
What are the two governance questions that agency AI creates?
When AI contributes to a client deliverable, two governance gaps open at once. The first is ownership. UK copyright law does not automatically protect AI-generated work with no identifiable human author, so ownership of an AI-assisted piece is determined by your contract, not by default. The second is disclosure. Clients paying for human craft react badly when they discover AI was involved.
Management consulting has already run into both. Harvard Business Review reported in late 2025 that consulting firms experimenting with AI in client engagements face live questions about whether AI-generated insights belong to the firm or the client, and that firms marketing AI-assisted work without disclosing the AI role carry genuine reputational exposure. The same dynamic applies to agencies. The deliverable is creative work, the client’s expectation is human expertise, and the commercial relationship depends on trust.
An ownership clause settles the first problem. A disclosure standard settles the second. Without both, you are one disappointed client away from a conversation you did not plan to have, about a decision that should have been made before the project started.
Why does agency reputational risk sit where regulated sectors put compliance?
Agencies are lightly regulated compared to law firms or financial advisers, which means there is no mandatory AI governance framework. That is an advantage until it backfires. In a regulated sector, undisclosed AI use creates a documented compliance breach. In an agency, it creates a reputational breach that is invisible until a client notices, and considerably harder to recover from once it has.
McKinsey’s 2025 Global AI Survey found that 88 per cent of organisations were regularly using AI in at least one business function, but only around a third had reached any kind of scaled integration. The majority are still in the pilot or ad-hoc phase, using tools without governance frameworks to match. Across sectors, the research on adoption found that lightly regulated industries could experiment with AI more rapidly, but faced higher reputational risk when something went wrong, precisely because no compliance framework had forced the governance conversation first.
For agencies, the gap matters differently to regulated sectors. A law firm that handles AI-assisted work without proper oversight answers to the Solicitors Regulation Authority. An accountancy practice answers to the ICAEW. You answer to your clients, and that is a more personal form of accountability than any regulator. The creative industry runs on relationships and repeat work, and a client who feels misled about how their work was produced is one who does not renew.
Where do the IP and disclosure questions surface in practice?
Both questions typically surface in three places. One is the contract your client signed before the work started. Another is the conversation when a client asks how a deliverable was produced. The third is the team’s daily tool use that no one has formally acknowledged. The last is where the exposure starts, because creative teams are using AI on live briefs right now, with or without a policy in place.
In the contract, the problem is what is absent. Many agency agreements do not have an AI-use clause, an ownership clause for AI-assisted work, or any language about what the agency is doing on the client’s behalf when it uses generative tools. That gap is fine in the straightforward case and a serious problem in the disputed one.
In the client conversation, the problem is precedent. If your team routinely produces AI-assisted work and nobody mentions it, then the first time a client asks you are either disclosing a practice that was already underway or explaining why it was not mentioned earlier. Both are harder conversations than an upfront disclosure standard would have been.
At team level, the British Chambers of Commerce reported in early 2026 that half of UK businesses were already using some form of AI. In agencies, where speed and output quality are both under pressure, the real rate of tool use is likely higher. The question is not whether your creative team uses AI on client work. The question is whether you have agreed what to tell clients about it.
When does client disclosure become non-negotiable?
The answer hinges on one thing. Whether the client’s expectation of human craft is material to what they agreed to pay. If a client commissioned original brand voice work and AI generated a substantial portion of the first draft, they have a legitimate interest in knowing. The question is commercial as much as ethical. A client who finds out later has been misled, and misled clients leave.
The SRA’s 2026 guidance on technology use in legal services sets a useful benchmark, even though agencies sit outside its scope. It requires that professionals maintain full responsibility for technology outputs and governance frameworks for AI in client-facing work. The underlying principle applies across any professional service. Where the client is paying for expert judgment, the role AI played in producing that output should be visible.
A practical agency standard does not need to be complicated. Three components are enough, and none requires legal review each time. One is a clause in the client contract covering AI use in deliverable production. Another is a default line the whole team uses when AI played a substantive role in the work. The third is an ownership clause that makes clear who can use AI-assisted outputs and how. Getting all three in place is a matter of deciding what the agency’s standard is, writing it down once, and following it consistently.
What else belongs in an agency AI governance conversation?
IP and disclosure are the first conversation, not the whole one. Once you have settled the contract clauses and the disclosure standard, three more areas need attention. One is how your team uses client data in AI tools. Another is what happens when a client disputes the ownership of AI-assisted work. The third is how your professional indemnity insurance covers AI-related failures. All three connect back to the contract, not the tool.
Client data in AI tools is the sharpest concern. If a creative team is feeding client briefs, brand guidelines, or proprietary research into a public LLM, that data may be retained and used in ways the client did not agree to. The ICO has published clear guidance on using generative AI in business contexts, including obligations around data protection when personal or confidential information appears in a prompt. Many agency tool setups do not meet that standard, because the policies were built for the team’s convenience, not the client’s data rights.
Professional indemnity is less obvious but equally important. Research from the Utah Bar’s analysis of professional liability in the era of generative AI found that insurers were scrutinising whether professionals had exercised reasonable care when relying on AI outputs. The same question applies to agency work. If an AI-assisted deliverable is found to be defective and the agency cannot show it was reviewed against the brief, the coverage question becomes a judgment call.
The broader point is that these three conversations, contract, data, and insurance, are more tractable now than after a client complaint. Set the standard when you have time to think, not when you need to defend yourself.
