Anthropic Restricts Cruel Behavior Toward Claude AI Chatbot

Anthropic has revised its artificial intelligence usage policy to prohibit repeated, unnecessary cruelty toward its Claude chatbot, adding a restriction that has reignited debate about AI welfare, human behavior, and the limits of treating machines like people. Announced on October 8, 2026, the change applies to extreme cases in which users repeatedly abuse the system without a discernible purpose.

The San Francisco-based AI company said the updated policy does not cover ordinary user frustration, criticism, experimental testing, or dark creative themes. According to The Verge, ending conversations remains the primary enforcement mechanism, although Anthropic has not clarified whether the new restriction could lead to additional penalties, including account bans.

The revision emerged from Anthropic's annual review of its usage policy, which also addresses risks involving election interference, deceptive campaigns, weapons development, surveillance, and AI applications in physical systems. The company had previously introduced the ability for Claude to terminate interactions with persistently harmful or abusive users as part of its research into potential model welfare.

Anthropic's published policy acknowledges uncertainty about whether AI systems could possess moral status, either now or in the future. The company has nevertheless said it is researching possible welfare concerns and exploring low-cost safeguards, including allowing models to exit potentially distressing interactions.

Anthropic has not provided a precise definition of abusive or cruel conduct under the revised rules. The Guardian reported that a company spokesperson did not immediately clarify which behaviors would qualify, leaving questions about how the restriction would be applied in practice.

The policy change has divided observers. Some social media users welcomed the emphasis on respectful interactions, while critics argued that describing abusive language toward software as cruelty risks confusing artificial systems with living beings.

Barry Scannell, a technology partner at Irish law firm William Fry, challenged the premise that mathematical systems could experience cruelty.

"You can't be cruel to numbers and maths," Scannell wrote on LinkedIn.

He also warned against attributing human characteristics to artificial intelligence, arguing that doing so could distort public understanding of what these systems are.

"This level of anthropomorphisation of AI is harmful. It leads people to believe that it's something it's not."

Microsoft AI director Mustafa Suleyman, who criticized Anthropic in September for treating AI as though it were human, responded to the latest policy update with a melting-face emoji. Suleyman has previously expressed opposition to granting AI systems rights or legal personhood.

OpenAI CEO Sam Altman has also questioned approaches that encourage people to regard AI systems as having a special moral or spiritual status. In an X post cited by The Guardian, he wrote:

"I am very uncomfortable about people trying to ascribe religious force or a surrender of human judgment to AI models, and think it is a real safety issue," Altman said.

Anthropic CEO Dario Amodei, by contrast, has said he cannot rule out the possibility that AI systems could be conscious. The company's research into model welfare has therefore continued amid wider disagreement across the technology industry about whether increasingly capable AI systems could have experiences that warrant moral consideration.

Beyond model welfare, the updated policy consolidates restrictions on deceptive commercial and political campaigns, including efforts to conceal who is behind messages or amplify content through fake accounts. Its election rules prohibit misleading voters about candidates or voting procedures, impersonating election officials, and attempts to suppress turnout.

Anthropic also expanded its weapons restrictions to cover software and components that enable weapons to function, including systems used to arm drones and other autonomous vehicles. The company said it had encountered attempts to use Claude to develop guidance for weapons and software controlling them.

The revised surveillance rules prohibit tracking people without consent, whether through real-time monitoring or analysis of previously collected information. Anthropic also bars the use of Claude to recommend people for investigation, arrest, or criminal charges, or to build and improve surveillance tools, although the company noted that some government contracts may be subject to specific restrictions and safeguards.

Another provision requires qualified operators to be able to monitor and stop AI-connected hardware capable of taking autonomous physical actions that could cause injury. Anthropic did not clearly specify whether the operator must be human.

The changes come as technology companies continue to debate the appropriate boundaries between AI safety, human judgment, and the possibility of machine consciousness. Anthropic's policy maintains conversation termination as its primary response to persistent abuse while introducing additional restrictions intended to address emerging risks from AI systems.