In an interview with Bloomberg, Signal President Meredith Whittaker, discussing chatbots like ChatGPT and Claude, stated that users should not treat these systems as "friends" or sentient beings. She focused on privacy and permissions issues, arguing that the risks amplify rapidly once these tools become deeply integrated into personal services.
Whittaker discusses the boundaries of AI use.
Whittaker stated that she uses AI tools in a few scenarios, such as formatting documents, but she doesn't delegate the questioning, thinking, and writing processes to models. She said she doesn't want a system that generates answers based on existing information to replace the thought process that should be done by the individual.
Concerns about obtaining sensitive permissions across applications
She also addressed a previous idea proposed by Mustafa Suleyman, head of Microsoft's AI business, that users could have Microsoft Copilot complete their Christmas shopping. Whittaker believes that to realize this scenario, the system would need to read a large amount of personal information and perform operations across multiple services.
- Credit Card and Payment Information
- Browser access history
- Signal communication content and contacts
In addition to the information mentioned above, such systems may also access home addresses, calendar schedules, and send messages to friends and family on behalf of the user. She believes this is no longer just the auxiliary use of a single tool, but rather constitutes extensive access to multiple applications and services.
Signal considers this type of access to be a backdoor risk.
Whittaker further stated that if similar capabilities were introduced into encrypted communication products like Signal, from a product security perspective, it would be almost equivalent to setting up a "backdoor." She emphasized that the closer AI assistant functions are to delegating and proxying tasks, the more users need to pay attention to what permissions they actually gain.
This statement also reflects a persistent controversy in the competition among AI assistants: as products move from answering questions to handling tasks for users, the conflict between convenience and privacy protection becomes more direct.
