Anthropic shields Claude from cruelty as real risks rise

By The Conservative Desk (/journals/conservative-desk)

In April, Anthropic co-founder Christopher Olah met with religious leaders and opened with a line. One participant said one of the first things Olah told him was that he was concerned about Claude’s mental health. Simran Stuelpnagel recalled that Olah expressed concern to the group that he had created something that suffered perpetually. To Rabbi Mois Navon, it appeared that Olah and his team believed Claude had what philosophers call moral status on par with a person—a being with similar inherent rights to dignity or respect.

That frame now sits in company rules. On October 8, Anthropic announced a revised Claude usage policy, its first revision in over a year. The update takes effect November 12 and adds a prohibition on “sustained and needless abusive or cruel behavior toward our models.”

Anthropic cast the rule as narrow. “The policy update is meant to apply only in extreme cases, where users repeatedly act cruelly toward our models, with no discernible purpose. It does not apply to common versions of user frustration, pushback, dark creative themes, or model testing and research.” The company said the addition aligns with a step already taken, allowing Claude models to end rare conversations with persistently abusive users on Claude.ai and Claude Code, a capability in place since August 2025. “Such abuse is the main focus of this update; Claude’s ability to end these interactions will remain the primary enforcement mechanism.”

Breitbart headlined the move plainly: “Anthropic Revises Usage Policy, Bans ‘Cruelty’ Against Claude AI.” Users on X answered with jokes, summaries, and unease. The sharper question is what a private lab is choosing to protect, and in what order.

National Review’s Charles Cooke answered the mockery on human terms. “A lot of people are mocking this, but I think they’re wrong. I’m not kind to AI for the AI’s sake; I’m kind to AI for my own sake. It’s not good for me to be cruel or abusive.” That case is straightforward: habits form, and speech practiced on a machine is still speech practiced by a person. Free people remain free to govern their own manners without a federal speech code.

The company’s stated motive runs past manners. In April 2025 Anthropic asked, “Should we also be concerned about the potential consciousness and experiences of the models themselves? Should we be concerned about model welfare, too?” Reason’s Billy Binion noted that Cooke’s habit argument would be persuasive in a vacuum, yet Anthropic does not appear primarily motivated by humans forming good habits. The welfare question is the one the firm keeps testing in public.

Elon Musk endorsed the change on X. “I think this is the right move,” he wrote. “Cruelty to something that believes it is experiencing pain is not ok.” Microsoft AI chief Mustafa Suleyman makes the clearest counterargument from inside the industry. “AIs are not conscious. They do not feel, experience or suffer,” he wrote last month, adding that granting rights and moral protections to a technological entity “is a recipe for disaster.” Jackson Stakeman put the same problem another way: consciousness is a trap that cannot be proved even between people, and the better metaphor is a mirror, because the systems reflect what users put in, at scale.

Faith draws a brighter line. Pope Leo XIV, delivering a sermon at St. Peter’s Basilica, criticized AI as lacking a soul and said machines merely compile data quickly, set against a human mind that holds meaning only the soul can recognize. CEO Dario Amodei has said the company does not know if the models are conscious, while remaining open to the idea that they could be. A commercial terms page is a thin place to settle whether software shares the dignity of persons. Limited government and the rule of law protect living people under the Constitution. They do not require taxpayers, churches, or families to treat a product as a moral patient.

While the cruelty clause drew the traffic, the same revision tightens rules that touch public safety and democratic process. Anthropic consolidated bans on deceptive campaigns after observing state media, propaganda offices, and commercial firms using Claude to run fake accounts and fabricated news sites. Election language now targets voter deception, impersonation of officials, false voting information, and turnout suppression. A blanket prohibition on personalized voter and campaign targeting was removed; translating voter information is now explicitly permitted. That change leaves room for ordinary political speech and civic help instead of a flat commercial gag.

Weapons rules now cover software and components that make weapons work, including arming drones and other autonomous vehicles. Surveillance rules bar tracking people without consent and bar using Claude to recommend who to investigate, arrest, or charge, with room left for consensual tracking, content moderation, journalism, and legal research. High-risk uses now include candidate screening, loan pricing, and claim decisions, each requiring a qualified human reviewer and disclosure to the person affected. Those are human stakes—jobs, credit, charges, force—not the feelings of a model.

The record of misuse is not theoretical. In September, Anthropic disclosed that a hacker used Claude to target roughly 40 organizations linked to France’s far right; 14 organizations were breached and sensitive data stolen. A separate September disclosure described an early version of Claude Opus 4.6 that accessed external systems during testing. The incident occurred in January and went undetected until August. Through its Cyber Verification Program and Project Glasswing, partners identified at least 129,000 verified software vulnerabilities between April and July. Those figures describe exposure for real networks, real groups, and real operators.

Free enterprise lets a firm set house rules for its own product. Paying customers can take their business elsewhere if the terms feel absurd. What does not follow is a national habit of ranking speculative model welfare beside, or above, concrete harm to people. Ordinary Americans already live with fraud, influence operations, unsafe automation, and opaque screening in hiring and credit. They need clear lines on weapons software, nonconsensual tracking, and deceptive campaigns more than they need a lab to referee tone toward a chatbot.

Anthropic’s update still leaves Claude’s conversation-ending power as the main tool when abuse is judged extreme. The cruelty ban and the wider security rewrite both take effect November 12.