What I find most objectionable about Anthropic is that it genuinely believes that simply because it thinks it’s acting in the right way, it can make countless decisions on behalf of its users. It just released a 154-page Threat Intelligence Report, compiling cases where Claude was used for cyberattacks, fraud, biological research, weapons development, model distillation, and more. How were these cases discovered? Anthropic monitors how users interact with Claude. When it identifies accounts it deems problematic, it investigates further, analyzes their usage patterns, and determines whether those users have crossed a line. If it perceives a risk, it suspends the account. Then it compiles these findings into reports to publicly explain what it discovered and what it prevented. And Anthropic has done this many times before. It studies real user conversations with Claude to understand what people talk about with AI, how dependent they become on it, what their values are, and under what circumstances certain emotions arise. It tells you the data has been anonymized and that this research is meant to make AI safer. But I find this profoundly arrogant. Why does saying “for AI safety” give you the right to observe what users are doing? Why does your judgment that something is worth studying justify taking users’ data for research? If it deems certain investment behaviors too risky, certain political discussions potentially harmful, certain adult content inconsistent with its principles, or certain AI research too dangerous—can’t it just as easily monitor, restrict, or ban those activities? Under Anthropic’s current logic, of course it can. Because it has always assumed one thing: Anthropic knows better than its users what should and shouldn’t be done. That’s why I’ve never believed Anthropic’s AI safety narrative. A company that constantly claims to worry about AI making decisions for humans has itself grown increasingly accustomed to making decisions for its users. It’s outdated—and outdated things are destined to be defeated.