Brands are demanding stronger AI safeguards in their contracts with ad agencies, setting up tension between marketing firms, which are racing to adopt generative AI, and clients, who are worried about all of the ways the technology could steer them wrong.

“Recently, we won three new pieces of business and in the [master service agreement] it says, ‘you’re not allowed to use AI of any kind, without prior authorization,’” said one independent ad agency CEO, who spoke with Ad Age on condition of anonymity to protect the identities of clients. “So, that even means they don’t want us to use AI to help work on concepts, not just anything that goes out the door.”

The agency is not alone, as these strict AI terms are becoming more common, according to ad industry leaders. Last year, the Association of National Advertisers updated its guidance to brands, advising them to include clauses about AI and consent in their agency deals. Now, those restrictions are being applied.

“There is this juxtaposition of agencies ramping up their AI knowledge and usage, and then clients clamping it down,” the agency CEO said.

Last year, tech platforms started rolling out more generative AI tools, following the popularity of OpenAI’s ChatGPT, the text-generating AI. Google, Meta, Microsoft, Amazon and others have been developing their own text- and image-generating tools, as well. Meanwhile, major ad agencies and media holding companies are developing services for brands based on these large language models.

However, there have been many high-profile stumbles with generative AI that have given some brands reasons to pause. In February, Google’s Gemini chatbot created embarrassing images when it was opened to the public. Last year, The New York Times sued OpenAI, and its top investor Microsoft, claiming ChatGPT’s large language model was trained on the publisher’s proprietary news archive. These thorny questions around copyrights and uncontrolled AI outputs are not new, but brands are just coming to grips with the technology.

Earlier this month, Under Armour ignited a debate in advertising after it released a commercial that used generative AI trained on its brand assets to create new visuals and voiceover of boxer Anthony Joshua, all without shooting new footage. Creatives questioned the ethics of repurposing old work.

More: Under Armour ad director addresses AI’s role

Still, AI has become a useful technology in most ad agencies, where media buyers use AI to analyze campaign strategies, and creatives use it to brainstorm. Brands are working with their top agency partners to train proprietary large language models suited to their specific brand identities.

“Every client meeting starts with this conversation,” said Robert Wrubel, managing director at Silverside AI, an AI lab within agency Pereira O’Dell. “It is always the initial question, just to make sure the systems and the processes we use are working within the standards and the sort of frameworks that the brands want to establish.”

Risk makers

Brands don’t want to mistakenly put out work that carries the imprimatur of another brand’s creative, Wrubel said. Brands also don’t want agency partners using AI in ways that could jeopardize their own intellectual property. For instance, if a brand’s customer information were fed into an AI system, that data could train a model that a competitor might use.

“How can we be sure that what we produce is going to be distributed and is appropriately managed so it doesn’t end up in another AI system?” Wrubel said.

High-tech ad agencies and brands are protecting their data by building closed systems. “We guarantee to brands and marketing teams that all of the work that we’re doing in generative AI is working in environments that are not intermingling with other systems or other models,” Wrubel said.

AI is becoming more ingrained in commercial work and internet marketing. Major brands, including Coca-Cola, Hyundai, Avocados From Mexico and others have used generative AI in their public-facing campaigns, giving consumers ways to generate personalized content. Advertisers also use generative AI in Google and Meta’s ad platforms, which are automated to create and target campaigns.

Brands need to adopt AI, but they also don’t want to inadvertently lose control of data and assets, according to one top tech executive at a major ad agency, who spoke on condition of anonymity. “What brands have not figured out is how do I protect my asset from someone else being able to monetize off of it,” the agency exec said. “So I think that’s why they’re putting that in the contracts. I’ve seen that in contracts.”

“So what we do is, we’ll build this walled garden,” the agency exec said, which means a brand stores its data in a closed system.

Careful to a fault

Brands could be too cautious, though, by telling agencies they need clearance for every usage of AI. There is a distinction between generative AI tools for creating ads and machine learning applications, which are widely used to analyze data, target ads and measure campaigns, according to Ashwini Karandikar, executive VP, media, tech and data, at 4A’s, the ad agency trade group.

“To suggest that the agency should disclose and obtain the advertiser’s approval for any such use serves only to unnecessarily complicate the contractual requirements when there is no practical benefit for the advertiser in doing so,” Karandikar said. “The real need is to understand where the risk is and develop guidelines that mitigate or ensure clarity of those risks.”

Some agencies are working with brands that adopt strict “no AI” policies, according to Chris Jacks, director of growth strategy at marketing agency HireInfluence. Contracts from Fortune 500 brands are coming with clauses that forbid the use of generative AI in the “creation of campaign deliverables,” Jacks told Ad Age.

AI is becoming a major factor in agency reviews, too, according to Greg Paull, principal at R3, a consulting firm. ‘We’ve got a couple of big reviews out there right now,” Paull said, “and in all of them, AI is part of the review, whereas a year ago nobody really thought to put that in.”

Contributing: Asa Hiken