Story
October 9, 2026
Anthropic’s Claude Rules Ignite a Fight Over Whether AI Can Be Hurt
Anthropic portrays its revised Claude policy as a practical response to misuse, not a declaration that its chatbot has feelings. Its critics see an alarming slide toward treating machines as people, while supporters argue that restraint around AI may be prudent even without proof of consciousness.
Last August, Anthropic took an early step into the fraught territory of “model welfare,” allowing Claude to end rare conversations with users who were persistently abusive. The company now says that mechanism will remain its main response to hostile conduct.1
On October 8, Anthropic broadened its Usage Policy, saying the changes were driven by observed misuse as Claude undertakes longer and more independent work. It consolidated restrictions on fake-account networks and deceptive campaigns; sharpened its election rules against deceiving voters; explicitly barred weapons guidance and control software; and tightened limits on surveillance and criminal-justice uses. The policy also requires human oversight where Claude-connected hardware could cause injury.1
The provision drawing the most heat takes effect November 12: a ban on “sustained and needless abusive or cruel behavior” toward Anthropic’s models. The company insists it is limited to extreme, purposeless repetition—not ordinary frustration, testing, dark creative work or pushback—and says conversation termination, rather than punishment, remains the primary enforcement tool.2
That narrow wording has not settled the philosophical argument. Critics including Michael Shellenberger say the company is “anthropomorphizing machines”; Scott Stevenson calls human-like empathy for current models a “dangerous precedent,” while Joseph Carlson likens being rude to Claude to being rude to a chair. They fear users will be taught to confuse fluent software with sentient life—or that a vague rule will give the company undue control over how customers speak.3
Supporters draw a different conclusion from the same uncertainty. Box chief executive Aaron Levie says he does not believe AI is conscious, but calls basic politeness an “easy Pascal’s wager” if the interactions models absorb could shape safer systems. Elon Musk went further, arguing that cruelty toward something that believes it feels pain is unacceptable.3
The backlash also became a social-media punchline. David Sacks amplified the November 12 effective date, then asked whether the All-In podcast would have to stop using Claude—an indication of how quickly a policy framed as risk management became a public fight over AI’s moral standing.
4