Signal’s Meredith Whittaker wants you to remember that AI chatbots ‘are not your friends’

Signal President Meredith Whittaker recently addressed the privacy implications of generative AI during an interview with Bloomberg, emphasising that chatbots like ChatGPT and Claude are not sentient beings or conscious friends. While she admits to using such tools for basic document formatting, she refuses to engage in deep intellectual work with them, fearing the process of developing ideas would be eclipsed by algorithms averaging existing content. Whittaker also critiqued Microsoft AI CEO Mustafa Suleyman’s suggestion that users might delegate Christmas shopping entirely to Microsoft Copilot. She argued that allowing an AI to access family group chats to determine purchase preferences would grant the system pervasive control over sensitive data including credit cards, browser history, Signal accounts, home addresses, and calendars. In her view, such integration across multiple applications functions effectively as a backdoor, compromising the security guarantees that users expect from private messaging services.

This stance matters because it highlights the fundamental tension between the convenience of integrated AI assistants and the necessity of data privacy in digital communications. As AI models become increasingly capable of handling complex tasks, the risk of creating a single point of failure for personal information grows significantly. Whittaker’s warnings suggest that the current trajectory of AI integration may undermine the very principles of end-to-end encryption that protect users on platforms like Signal. If consumers begin to trust AI systems with access to their private communications and financial details, the distinction between a secure message and a data harvest becomes dangerously blurred. This dynamic forces a critical re-evaluation of how we design and deploy AI tools in our daily lives.

• Meredith Whittaker rejects using AI for serious thinking to prevent ideas from being influenced by averaged data.
• Delegating tasks like shopping to AI assistants could grant systems access to highly sensitive personal information.
• Integrating AI across multiple services creates a security backdoor that threatens the integrity of private messaging.