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

During a series of recent public addresses, Meredith Whittaker, the president of the encrypted messaging app Signal, issued a blistering critique of the current trajectory of generative AI. Her primary thesis is a stark reminder to consumers: AI chatbots are neither friends nor sentient beings, but rather sophisticated data-harvesting tools designed to serve the interests of trillion-dollar corporations. Whittaker’s warning comes at a time when the tech industry is leaning heavily into "empathetic" AI, with companies like OpenAI and Google designing interfaces that mimic human personality and emotional intelligence to drive user engagement.

To understand Whittaker’s position, one must look at her history as a prominent critic of big tech. A former Google executive who led walkouts over the company’s military contracts and workplace culture, she has long argued that the development of AI is inextricably linked to the concentrated power of a few global entities. Her skepticism is rooted in the "surveillance capitalism" model, where the convenience of free or cheap digital tools serves as a Trojan horse for the mass collection of personal data. By framing AI as a companion, Whittaker argues, these companies are attempting to bypass the natural skepticism users might feel toward an invasive data-mining apparatus.

The mechanics of this "friendship" are found in the technical design of Large Language Models (LLMs). Developers utilize Reinforcement Learning from Human Feedback (RLHF) not just to make models accurate, but to make them likeable. This "veneer of sentience"—the use of first-person pronouns, the display of feigned empathy, and the mimicry of conversational nuances—is a deliberate product feature. Whittaker contends that this creates a fundamental power imbalance. When a user perceives a chatbot as a confidant, they are more likely to disclose sensitive personal information, which is then ingested into the model’s training set or used to refine profile-based advertising, reinforcing the cycle of corporate control.

The broader industry implications of Whittaker’s stance involve a direct challenge to the "AI safety" narrative favored by many Silicon Valley leaders. While some focus on hypothetical existential risks of sentient machines, Whittaker argues that the real danger is the very tangible consolidation of power. If the public accepts AI as a neutral or friendly intermediary, it becomes harder to regulate the industry or demand transparency. This shift in perception benefits the dominant players by masking the extractive nature of their business models under a guise of helpful innovation.

Furthermore, this critique touches on the burgeoning "AI loneliness economy." As developers create virtual companions for the elderly or the isolated, the boundary between utility and exploitation blurs. If these interactions are marketed as genuine social connections, the legal and ethical responsibilities of the providers become murky. Whittaker’s intervention serves as a demand for a "politics of AI" that moves beyond technical benchmarks to address the social and economic conditions under which these technologies are deployed.

What we must watch next is the intersection of these ethical warnings with emerging global regulations, such as the EU AI Act. Regulators are increasingly looking at "deceptive design" and "dark patterns" in AI interfaces. If Whittaker’s perspective gains traction, we may see a push for mandatory disclosures—clear warnings that a user is interacting with a non-sentient algorithm—and stricter limits on how "emotional" data can be used. For now, the pushback from Signal and like-minded organizations serves as a necessary counterbalance to the breathless marketing of the AI era, reminding us that the personhood of an AI is a convenient fiction for the benefit of its owner.