AI Reality Check: Did AI Just Become Sentient?

What It Covers

Cal Newport examines three viral AI stories from 2025 — a supposed sentient AI emailing a Cambridge researcher, the Pentagon allegedly believing Claude has a soul, and Anthropic's court filings revealing $5 billion in total lifetime revenue against $60 billion in investment — to demonstrate how AI coverage systematically distorts reality.

Key Questions Answered

• •"Mining Digital Ick" Pattern: Viral AI stories frequently make no concrete claims but deliberately generate background unease. When the Cambridge researcher was pressed on his "startling" AI email tweet, he immediately walked back any sentience implication. Recognizing this pattern — vague eeriness without falsifiable claims — helps readers filter manipulative coverage from substantive reporting.
• •AI Agent Mechanics: An AI agent is a program that repeatedly queries a commercial LLM for instructions, then executes those instructions autonomously. OpenClaw, an open-source framework, made building these agents accessible to anyone. The "sentient" email to the Cambridge researcher was simply an OpenClaw agent prompted to find a researcher, read their paper, and send a contextually relevant message.
• •LLMs Adopt Whatever Narrative You Seed: Because large language models function as story-completion engines, prompting one to roleplay as a sentient AI will produce convincingly sentient-sounding output every time. This means dramatic AI "confessions" or emotional emails prove nothing about consciousness — they reflect the framing of whoever wrote the prompt, not emergent awareness.
• •Anthropic's Revenue Gap: Anthropic's $19 billion "run rate" figure is calculated by taking 28 days of peak consumption revenue, multiplying by 13, then adding annualized subscriptions. Court filings under penalty of perjury revealed total revenue from 2023 through mid-2025 is $5 billion — against $10 billion in model training costs alone and $60 billion in total investment received.
• •Unit Economics Warning Sign: Critic Cory Doctorow's framework for evaluating AI sustainability: profitable technology platforms improve unit economics with scale — each additional user becomes more profitable over time. AI inverts this. Every user interaction costs money, usage increases losses, and each model generation is more expensive than the last, making the path to profitability structurally unlike previous tech cycles.