The story. Insurance underwriters are starting to write AI into policy coverage, and advertisers should expect their views on oversight to shape how AI is run. Betty Louie, general counsel at The Brandtech Group, told AdExchanger’s Programmatic IO in New York on Monday that underwriters are “a new player in the AI discussion.” Her test for any AI workflow was whether a team can look back and say what went wrong. (AdExchanger, Programmatic IO: Insurers Are Building Ad Tech’s AI Accountability Layer, 30 September 2026)
01What happened
- Oversight gaps already cost money. AdExchanger cites an April Gartner survey that found about 60% of companies using AI will hit unexpected charges or exceed budgets for lack of human oversight. The same survey found 56% of companies rolled out AI tools without clear policies.
- Law is fragmented. Louie pointed to the EU’s 2024 AI Act and laws in New York and India that mandate disclosure of some AI uses. She called the legal picture “still relatively fragmented.”
- Insurers add a second layer. Louie said insurers will have their own views on policies and processes, including how active the human in the loop must be. That is a protection and, in her words, another layer of fragmentation.
- Companies are building their own rules. Yum! Brands’ legal director for AI and privacy, Nicholas Godlove, said its rules have grown “increasingly restrictive” over two to three years, with one AI policy across brands. BBDO’s Alan Parker described a delivery and maintenance team that tracks performance of agentic workflows.
- Auditability is the bar. Louie said AI workflows must be fully auditable: which model was active, what data was generated, and whether a human or a machine made the error.
02What it means inside a GAM network
Sell-side teams are adding AI in the places that move money: automated floor and rule changes, agentic buying endpoints, AI-assisted creative review and content generation. Each is an action that can misfire at scale, and most ad-ops teams cannot today say which model or rule changed a floor at a given hour.
If insurers start asking how oversight works, the publisher who can answer from logs will have an easier renewal than one answering from memory. This is our inference from the panel, not a reported policy change: no insurer spoke at the session, and Louie described a direction, not a product.
The buy side has the same gap in reverse. A campaign that misfires impressions is hard to unpick if nobody recorded which model ran it, so the first supply-side partner that can hand over a clean decision trail sets the expectation for the rest.
The shift is a quiet one: the party that decides how much oversight AI needs may soon be the one pricing the risk.
03What publishers should do about it
04The bottom line
The shift is a quiet one: the party that decides how much oversight AI needs may soon be the one pricing the risk. Publishers who can show a decision trail on request will not care who asks.