Who answers when AI rewrites approved creative

A bookstore's holiday campaign went live with mangled text and a swapped product photo. Nobody on the team had made those edits: Meta's ad AI changed the creative after the assets had been approved and launched. The photographer whose work the algorithm rewrote found out from other people's messages about the «AI slop» showing up in their feeds.

What made the story travel was not the glitch but the question the team could not answer: who owns this. Greg Jarboe puts it bluntly — AI is not creating an accountability crisis, it is exposing the gaps that were already in the process. The tooling made an unrequested, unflagged change to approved assets, and nobody's job was to compare the live creative against the signed-off file. The column ran on Search Engine Journal.

Guy Hanson of Validity breaks a campaign into stages and shows exactly where ownership falls through:

• strategy — still firmly human;
• building — the territory AI has almost fully colonised;
• approval and handoff to live — the stage increasingly slipping between teams;
• in large teams blame disperses across departments and sign-offs; in small ones a single person carries strategy, execution and QA at once.

Validity's State of Email 2026 survey confirms the shift in priorities: 35% of teams now hire for AI and ML skills, 27% for automation and workflow development, 15% for compliance and data privacy expertise. Design and HTML, the single biggest requirement back in 2023, are down to 14%.

If there is no human name next to an AI agent, you don't have a process — you have hope. Name an owner for every agent that touches customers: that person answers for the permissions granted and the data used. Define what the system is allowed to change after sign-off, and audit the live creative against the approved file within 24 hours. And stop treating the AI budget and the compliance budget as competing lines — they are one line. Accountability cannot be automated; it can only be assigned.