Signal chief warns AI assistants could create data backdoors

Meredith Whittaker, the president of the encrypted messaging service Signal, has issued a warning regarding the nature of artificial intelligence chatbots, asserting that they are not conscious or sentient. Speaking in an interview about policy, privacy, and the Signal platform, Whittaker cautioned users against treating these systems as human-like entities. “These are not your friends. These are not conscious beings. These are not sentient interlocutors,” Whittaker said, addressing the privacy implications of chatbots such as ChatGPT and Claude.

While Whittaker acknowledged using AI tools for minor tasks, such as formatting a document here and there, she explained that she strictly avoids using them for her core intellectual work. She stated that she does not ask chatbots questions because she is serious about her own thinking and writing. Whittaker expressed concern that using AI tools to work through ideas can foreclose or eclipse the thinking process, as these systems merely average existing information that is already out there.

Whittaker also raised privacy concerns regarding the integration of AI assistants into personal applications, specifically targeting Microsoft Copilot. Her critique followed a prediction by Microsoft AI CEO Mustafa Suleyman that users could let Copilot handle all of their Christmas shopping this year. Whittaker argued that for Copilot to perform such tasks, it would require access to a user’s credit cards, browser history, home address, calendar, and messaging applications like Signal, as well as the ability to message siblings on the user’s behalf.

Whittaker characterized this scenario as eavesdropping on the family group chat to determine what gifts people want. She warned that granting an AI assistant this level of integration creates “a system with very pervasive access” across multiple applications and services. In the context of Signal, which is designed for secure, private communication, Whittaker asserted that such integration would constitute “a kind of a backdoor” into personal user data. She emphasized that allowing an AI assistant to have this level of access across multiple applications and services would undermine the privacy guarantees of secure communication platforms.

Why it matters

As AI assistants become deeply integrated into personal productivity tools, Whittaker’s critique highlights the tension between convenience and the security risks of granting AI pervasive access to private communications and data.