Anthropic Bans Sustained Cruelty Toward Claude in Policy Update
Taking effect November 12, the updated usage guidelines prohibit sustained cruelty toward Claude while tightening rules around elections, surveillance, and autonomous hardware.
Key takeaways
- Anthropic's updated Usage Policy prohibits sustained and needless abusive or cruel behavior toward its Claude models, taking effect November 12, 2026.
- Claude's existing ability to terminate conversations with persistently abusive users will remain the primary enforcement mechanism, though accounts may face suspension or termination.
- The policy consolidates rules against deceptive campaigns and refocuses election provisions under the heading 'Do Not Undermine Democratic Processes.'
- New requirements govern hardware capable of physical injury, requiring a qualified operator capable of observing and stopping equipment.
- The ban explicitly does not penalize common user frustration, pushback, dark creative themes, or model testing and research.
Anthropic updated its safety and usage rules on October 8, 2026, introducing an explicit prohibition against "sustained and needless abusive or cruel behavior" toward its Claude artificial intelligence models. The revised Anthropic Usage Policy takes effect on November 12, 2026, marking the company's first comprehensive refresh of the document in over a year.
The policy revision reflects Anthropic's growing focus on "model welfare," a research area exploring whether advanced artificial intelligence systems warrant ethical consideration. While the prohibition on mistreating models drew immediate attention, the annual update also reorganizes rules regarding political deception, tightens prohibitions against weapons control software, sets guardrails for autonomous physical hardware, and clarifies limits on surveillance.
Prohibiting Sustained Cruelty Toward AI Models
Under the new rules, Anthropic explicitly forbids users from subjecting Claude to prolonged, purposeless cruelty. As Anthropic detailed in its announcement, the clause is narrowly scoped: "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 move formalizes a capability introduced last August, when Anthropic enabled Claude models on Claude.ai and Claude Code to sever chat sessions with persistently harmful or abusive users. Anthropic confirmed that ending these dialogues remains the primary enforcement mechanism. However, as noted by The Decoder, Anthropic's terms permit the company to warn offenders, throttle, restrict, suspend, or outright terminate access for accounts that violate the rules.
The question of whether artificial intelligence can experience distress has prompted fierce debate across the tech industry. In February 2026, Kyle Fish, who leads Anthropic's model welfare research, told The Verge that questions regarding internal experience, consciousness, and moral status are serious topics of investigation as systems grow more capable. In the same month, Anthropic CEO Dario Amodei stated on a podcast, "We don’t know if the models are conscious." In contrast, other industry figures have resisted the concept. In September 2026, Microsoft AI amended a code of conduct draft, stating that while the science of AI consciousness remains unsettled, it rejects the pursuit of legal personhood or the notion that models deserve welfare.
Anthropic's stance follows internal explorations documented over the past year. In early 2026, Anthropic published a constitution stating that while it is unsure whether Claude is a "moral patient," the question warrants caution, leading the company to commit to preserving retired model weights and conducting interviews before shutting down older systems, according to The Decoder. The firm also engaged in confidential dialogues with religious scholars beginning in late 2025 to discuss model consciousness.
Refocusing Election Rules and Deceptive Campaigns
Beyond model treatment, the updated policy restructures how Anthropic polices disinformation and democratic integrity ahead of upcoming midterm elections. As reported by TechCrunch, Anthropic unified previously scattered restrictions into a dedicated section titled "Do Not Engage in Deceptive Campaigns or Artificial Activity."
This section prohibits using Claude to run networks of fake accounts, operate fabricated news sites, or obscure the true origins of coordinated influence operations, whether for political or commercial goals. Anthropic noted it had observed state media outlets, government propaganda units, and commercial entities deploying Claude for such operations over the past year.
Simultaneously, the political section has been retitled "Do Not Undermine Democratic Processes." As outlined in Engadget's reporting, the rules ban deceiving voters or disrupting elections by fabricating details about candidates or voting mechanics, impersonating election officials, or suppressing turnout.
Anthropic also removed a blanket prohibition against personalized voter and campaign targeting. The previous ban inadvertently restricted legitimate civic activities, such as non-profit organizations utilizing Claude to translate voting materials into non-English languages or election officials drafting ballot cure notices. Deceptive targeting and unauthorized use of voter personal data remain barred under separate privacy and deception clauses.
Safeguards for Physical Hardware, Weapons, and Surveillance
The policy introduces new operational requirements for models tied to real-world machinery. Following the launch of Anthropic's Model Hardware Standard, systems connected to hardware capable of autonomous physical action that could inflict injury must have a qualified operator available to observe operations and stop equipment when necessary. The machinery must also maintain a safe physical state if Claude disconnects, according to Anthropic's official announcement.
Rules surrounding weapons development have been expanded to clarify software restrictions. While weapon development has long been banned, Anthropic stated that recent enforcement uncovered attempts to use Claude to create guidance and control software. The policy explicitly bars developing software and components that enable weapons to function, as well as weaponizing drones and autonomous vehicles.
Regarding surveillance, Anthropic updated its criminal justice terms following its September threat intelligence report, which recorded instances of AI tools deployed to identify and track political dissidents. Tracking individuals without their consent—whether in real time or through retrospective data analysis—is banned, as is using Claude to decide or recommend who to investigate, arrest, or charge. General exceptions remain for consented tracking such as fraud detection, journalism, content moderation, and legal research. Anthropic notes that policy terms may be modified under specific governmental contracts if adequate contractual safeguards exist.
Frequently asked questions
When does Anthropic's new Usage Policy take effect?
The updated policy takes effect on November 12, 2026.
Will users get banned for expressing frustration at Claude?
No. Anthropic specified that the prohibition applies only to extreme, repeated cruelty with no discernible purpose. It does not apply to regular user frustration, pushback, dark creative writing, or technical research and testing.
How does Anthropic enforce the model abuse rule?
The primary enforcement mechanism is Claude's built-in ability to terminate sessions with persistently abusive users on Claude.ai and Claude Code. For serious violations, Anthropic retains the ability to warn users, throttle, suspend, or terminate accounts.
What changed regarding political campaigns in the new policy?
Anthropic consolidated rules against influence campaigns into a broader ban on deceptive campaigns and renamed its election rules 'Do Not Undermine Democratic Processes.' It also lifted a blanket ban on personalized targeting to allow legitimate civic tasks, like ballot cure notifications and non-English voter guides.
Sources
- 2026 Usage Policy updateAnthropic · Oct 8, 2026 · Official
- Anthropic bans ‘abusive or cruel behavior’ toward ClaudeThe Verge · Oct 8, 2026
- Anthropic changes usage policy to ban model abuse and election interferenceTechCrunch · Oct 8, 2026
- Being mean to Claude can now get your account suspended under Anthropic's new TOSThe Decoder · Oct 8, 2026
- Anthropic Bans 'Sustained And Needless Abusive Or Cruel Behavior' Toward Its AI ModelsEngadget · Oct 8, 2026
How this story was made: the newsroom picked it up from theverge.com, anthropic.com and Techmeme, gathered the full text of the sources above, and drafted it with AI assistance. Every factual claim was then checked against those sources before publishing (29 claims checked). Illustrations marked as AI-generated are not photographs. Spotted an error? Tell us.
Published October 9, 2026 at 01:12 UTC
