adtestbench Higgsfield
EN

AI news · AI agents · privacy controls

14 of 29 AI products document an ad data opt-out, paper finds

Researchers counted 788 privacy settings across 30 AI assistants and agent tools, and found no field for privacy choices in the protocols agents use.

By Alexandre S. , 02:30 UTC

Just 14 of the 29 AI assistants and agent platforms where the question applies document a way to stop a person’s data being sold or shared for advertising, according to a paper revised on arXiv on 22 October 2026, PPC Land reports. Its seven authors, from Carnegie Mellon, Stanford, UC San Diego, Villanova and Wesleyan, catalogued 788 privacy settings across 30 products made by 18 companies.

Retro-futurist illustration: a striped sunset over a grid horizon under a starry sky, with a shield bearing a padlock standing on the horizon.
Drawn by adtestbench from “Privacy Preference Controls for Agentic Systems”,

Other switches are rarer still. An opt-out from ad personalisation shows up in 10 of 23 products, a way to refuse human checks of conversations in 6 of 30. Under half of the settings state what they are set to by default. The authors also note that Model Context Protocol and Agent2Agent, the shared rules agents use to connect to tools and to each other, have no field for privacy preferences, and they publish eight open-source prototypes that add one.

The plans marketers pay for state some of this on their own cards. On our 31 October check, ChatGPT’s Go plan card says “This plan may include ads”, while Claude’s Team seat lists no model training on your content by default.

For a brand whose team pastes campaign plans or customer notes into these assistants, the count means 15 of the 29 products document no way to keep that data out of ad sharing, and the authors did not check whether the 14 that do actually honour it.