Signal president Meredith Whittaker has issued a stark warning against the anthropomorphisation of AI chatbots, arguing that marketing narratives positioning large language models as sentient companions create significant liability and safety risks for enterprises deploying these systems.
Speaking at a technology conference, Whittaker—a prominent privacy advocate and former Google AI researcher—challenged the industry’s tendency to describe chatbots using friendship and relationship terminology, noting this framing obscures the commercial surveillance infrastructure underlying most consumer AI products.
The intervention comes as enterprises accelerate AI integration whilst grappling with employee and customer expectations shaped by consumer products marketed with emotionally resonant language. Major providers including OpenAI, Anthropic, and Google have deployed chatbots described using terms suggesting personality, helpfulness, and even companionship—language Whittaker argues fundamentally misrepresents the technology’s capabilities and commercial purpose.
“These are not friends. These are not sentient beings. These are products built by corporations with specific business models,” Whittaker stated, according to TechCrunch AI. Her comments directly address a growing concern amongst enterprise risk officers: that user misunderstanding of AI capabilities creates liability exposure when employees or customers rely on chatbot outputs for consequential decisions.
The business implications extend beyond liability. Organisations deploying AI systems face a perception gap where users trained on consumer chatbots expect human-like understanding and judgement that current systems cannot provide. This mismatch affects everything from customer service automation to internal productivity tools, where overreliance on AI outputs can compromise quality and accuracy.
Financial services firms have proven particularly cautious. Morgan Stanley’s deployment of GPT-4 to financial advisers, announced in 2023, included extensive guardrails and training specifically addressing the limitations of AI-generated advice. The insurance sector faces similar challenges, with underwriters reporting pressure to integrate AI whilst managing regulatory requirements that assume human judgement.
Whittaker’s position carries particular weight given Signal’s role as a privacy-focused messaging platform serving over 40 million users, including journalists, activists, and security-conscious enterprises. Unlike competitors monetising through advertising or data collection, Signal’s nonprofit structure provides distance from the commercial incentives driving anthropomorphic marketing.
The critique also highlights a strategic divide in the AI industry. Whilst consumer-focused providers benefit from emotional engagement that drives usage and retention, enterprise vendors face procurement committees demanding clear documentation of capabilities, limitations, and failure modes—documentation that conflicts with friendship narratives.
This tension manifests in contracting language. Enterprise AI agreements increasingly include explicit disclaimers about system limitations, liability caps for erroneous outputs, and requirements for human oversight—legal protections that implicitly acknowledge the gap between marketing perception and technical reality.
The regulatory environment is evolving to address these concerns. The EU’s AI Act, entering force in stages through 2026, includes transparency requirements for AI systems that interact with humans, mandating disclosure when users are engaging with automated systems rather than people. Similar provisions appear in proposed US state-level legislation.
For enterprises, Whittaker’s warning suggests three immediate considerations: auditing internal AI communications to ensure accurate representation of capabilities; implementing training programmes that address employee misconceptions shaped by consumer AI marketing; and reviewing liability exposure from customer interactions with AI systems marketed or perceived as providing expert judgement.
The anthropomorphisation debate will likely intensify as AI providers pursue stickier user engagement whilst enterprises demand more precise capability descriptions. Organisations should monitor regulatory developments around AI transparency requirements and prepare for potential liability frameworks that distinguish between systems marketed as tools versus those positioned as advisers or companions.
Whittaker’s intervention reframes AI deployment as a communication challenge as much as a technical one, with significant implications for how enterprises manage both internal adoption and customer-facing applications.
