Privacy expert warns AI agents can expose people who never agreed to use them

Declining to use an AI assistant may not keep someone’s information out of it. In an October 19 Q&A published by Syracuse University, privacy-law expert Alexis Shore Ingber warned that a friend’s decision to use an agent can expose other people’s communications—and leave those people with few practical or legal options.

Ingber, an assistant professor of communications at Syracuse’s Newhouse School, focused on Meta’s Muse agent. Meta says Muse handles private information carefully and lets users decide how much access to grant. Her objection concerns a different person: the friend, contact or conversation partner whose information becomes accessible through someone else’s choice.

Remembering a friend is also collecting about them

Meta advertises Muse’s ability to remember details such as friends’ dietary restrictions when helping send invitations. That example captures the overlap between useful assistance and information about someone other than the account holder. The remembered detail helps the user complete a task, but it concerns a friend.

Ingber identifies two differences between AI agents and familiar ways people expose one another online, such as tagging a photo. The first is scale. She describes Muse as an all-encompassing assistant and says it does not let users shield particular interactions or people. In her account, the exposure extends beyond a single item deliberately shared.

The second is inference: drawing conclusions from collected information rather than simply storing it. Ingber points to judgments about what content someone wants to see or which products they might buy. She warns that agents can draw conclusions about other people they encounter, too, not just the person who signed up.

Permission from one person leaves another outside the decision

Asked what an exposed friend can realistically do, Ingber says there is not much beyond asking the user not to share personal information. She says such requests can work, citing her own empirical findings. But that approach places the burden on an interpersonal relationship rather than on the platform collecting the information.

Her legal criticism follows the same divide. Consent-based privacy law, she explains, assumes a company asks the person whose data it collects for permission. When another person supplies that information, the affected individual lacks that moment to accept or refuse. Ingber also says proving the level of harm required by privacy law would be difficult.

Platforms have the responsibility to protect people who do not consent to data collection by default.

Alexis Shore Ingber, in Syracuse University’s October 19 Q&A

A veto for the other participant

Ingber does not argue that useful assistants and privacy protections cannot coexist. She acknowledges the difficulty when helping one user requires access to other people’s communications. Her proposed controls would give those other participants a direct say, rather than relying only on the agent user’s settings or a request between friends.

• Send a consent request when another person wants to use an AI chatbot that may need access to someone’s communications.
• Let people forbid other users from giving an AI chatbot access to their communications.

Her preferred default is opt-in access, not opt-out. These are proposals for how platforms should handle shared communications, not protections she says Muse already provides. The distinction is central to her argument: allowing an account holder to control an assistant’s access does not give everyone in that person’s conversations an equivalent choice.