Signal CEO Warns Against Overreliance on AI Chatbots

Meredith Whittaker, CEO of Signal, has issued a warning against users placing excessive trust in AI chatbots. In an interview with Bloomberg, Whittaker firmly stated, “These are not your friends. They are not sentient beings. They are not conversational partners with agency,” pinpointing the potential dangers of fostering undue intimacy or trust in generative AI tools like ChatGPT and Claude.

Whittaker acknowledged that she occasionally uses AI tools “to format documents,” but she emphasized, “I do not ask AI questions. I take my thinking and writing seriously. I don’t want the process of developing ideas to be closed off or overshadowed by the responses of a system that averages existing information.” Her comments reflect a clear opposition to outsourcing cognitive processes to AI systems.

Criticism of Microsoft Copilot Use Cases

A key focus of the interview was a prediction by Mustafa Suleyman, CEO of Microsoft AI, who suggested that by the end of the year, users would be able to delegate all their Christmas shopping to Microsoft Copilot.

Whittaker scrutinized this scenario in detail. She pointed out that for Copilot to eavesdrop on family group chats and determine gift preferences, the system would require broad access to “credit cards, browsers, Signal, the ability to send proxy messages to siblings, home addresses, and calendars.”

“What you’re describing is a system that has extremely broad access across multiple applications and services,” Whittaker remarked. “In the context of Signal, that would constitute a kind of backdoor.”

The Nature of Privacy and Trust

At the core of Whittaker’s statements lies a warning against projecting human attributes onto AI chatbots. As Signal’s CEO, she has championed privacy-focused design centered on end-to-end encryption. With prior experience at Google in AI ethics and public policy, she possesses deep insights into the evolving relationship between technology and humanity.

Her assertion that “AI chatbots do not have consciousness or emotions” may appear self-evident, but psychological studies show that people are prone to developing empathy and trust toward the fluent responses generated by language models. This phenomenon raises concerns, especially as AI-driven counseling and mental health support services emerge, blurring the lines further.

Whittaker’s cautionary remarks highlight the tension between the convenience offered by AI and the potential erosion of human autonomy and privacy. Her concerns extend beyond the risk of data breaches to the possibility that outsourcing thought processes to AI could undermine critical thinking and creativity.

Implications for the Industry

Whittaker’s comments serve as a sobering critique of the growing fervor surrounding the AI industry. Major tech companies like Microsoft, Google, Anthropic, and OpenAI are accelerating the development of “agent AI” systems designed to perform tasks on behalf of users. Suleyman’s vision of automating Christmas shopping is a prime example of this trend.

However, as Whittaker has pointed out, scenarios where such agents access family group chats, personal calendars, credit card information, and more significantly expand the traditional permissions model for apps. These developments could necessitate levels of access that current operating systems and application frameworks are not built to handle.

For instance, in a messaging app like Signal, which prides itself on end-to-end encryption, an AI agent reading messages and acting on that information to make purchases effectively nullifies the confidentiality of communications. This is why Whittaker likened such functionality to a “backdoor.” Unlike traditional privilege escalation vulnerabilities like the recently disclosed Microsoft Defender issue “RoguePlanet,” this represents a qualitatively new security challenge.

How Users Should Approach AI

Whittaker’s statements offer valuable guidance for all AI tool users. While AI chatbots can be useful tools, they are nothing more than that. Before sharing personal information, users must carefully consider how that data will be used and who will have access to it.

Her stance of “not outsourcing one’s thought process” provides a model for intellectual integrity in the age of AI. Instead of accepting AI-generated information at face value, users should critically evaluate it and maintain the habit of thinking independently.

From a privacy perspective, the fundamental principle of not disclosing overly personal information to AI chatbots is reaffirmed. Sensitive data—such as family or friends’ private information, financial details, or location information—should be handled with utmost care.

Editorial Opinion

In the short term, Whittaker’s statements may place a check on the marketing of AI agent services. As companies like Microsoft emphasize the message of “AI can do everything for you,” criticism rooted in privacy and trust concerns will likely prompt more cautious consumer behavior. Signal users, in particular, might become more wary of AI agents.

From a long-term perspective, this discussion could lead to a reevaluation of the fundamental design of permissions for AI systems. The current application-level permission models are ill-suited for scenarios in which AI agents autonomously operate across multiple services. The industry may need to adopt new authorization frameworks and principles of privacy by design. Additionally, as advancements like those demonstrated in Subquadratic’s breakthrough in overcoming Transformer limitations improve the efficiency of AI models, such systems will likely become even more deeply integrated into daily life.

References

• Signal’s Meredith Whittaker wants you to remember that AI chatbots ‘are not your friends’ - TechCrunch — Published on 2026-06-20