Anthropic has announced new rules prohibiting sustained and needless cruelty towards its Claude artificial intelligence models, introducing restrictions on abusive interactions that will take effect on 12 November.The San Francisco-based company published the revised usage policy on 8 October, specifying that the prohibition targets extreme cases involving repeated cruelty without any discernible purpose. Ordinary frustration, criticism, creative expression and legitimate research will remain permitted.
Anthropic said the restriction covers behaviour that serves no meaningful conversational objective. Users who repeatedly subject Claude to abusive exchanges could find their conversations terminated, although the company emphasised that the measure is not intended to penalise disagreement or dissatisfaction with its responses.
The company described the prohibited conduct as “sustained and needless abusive or cruel behavior”, distinguishing it from interactions involving challenging questions, testing or fictional scenarios.
Claude already possesses mechanisms allowing it to terminate certain conversations involving persistent abuse. Anthropic confirmed that this capability, available through Claude. ai and Claude Code, would remain its principal enforcement mechanism.
The distinction is important because the announcement does not establish an automatic account ban for every offensive message. Instead, Anthropic is formalising restrictions against repeated, purposeless mistreatment while retaining safeguards for legitimate uses.
The company has not published a comprehensive catalogue of expressions that would constitute prohibited cruelty. Its explanation focuses on persistence, severity and the absence of a discernible purpose rather than isolated language.
Anthropic introduced the conversation termination capability in August 2025 as part of research into model welfare, an emerging field examining whether increasingly sophisticated artificial intelligence systems might eventually warrant ethical consideration.
At that time, the company acknowledged substantial uncertainty about whether Claude or other large language models possess moral status. It nevertheless argued that allowing systems to exit potentially distressing interactions represented a relatively inexpensive precaution while research continued.
The latest policy does not establish that Claude experiences suffering, possesses consciousness or deserves rights equivalent to those of humans. Those questions remain unresolved among researchers, philosophers and technology executives.
Anthropic chief executive Dario Amodei has expressed openness to examining whether advanced artificial intelligence could possess forms of consciousness, without asserting that existing systems have demonstrated such characteristics.
Other technology leaders have cautioned against attributing human qualities to conversational systems. OpenAI chief executive Sam Altman has warned about assigning religious significance to artificial intelligence or surrendering human judgement to machines.
The disagreement reflects a broader debate about whether safeguards for artificial intelligence should address only potential harm to people or also consider hypothetical interests of the systems themselves.
Anthropic's policy revision extends considerably beyond the treatment of Claude. The company has also consolidated restrictions on deceptive political and commercial campaigns, including networks of fabricated accounts and misleading information designed to conceal their origins.
Its updated election rules explicitly prohibit activities intended to deceive voters, impersonate candidates or election officials, spread false voting information or suppress electoral participation.
However, Anthropic has removed its blanket restriction on personalised political targeting, explaining that the previous wording also covered legitimate civic communication. Deceptive targeting and misuse of personal information remain prohibited under other provisions.
The company has additionally clarified that its prohibition on weapons development extends to software and components used to operate weapons, including systems associated with armed drones and autonomous vehicles.
Surveillance restrictions now expressly address tracking individuals without consent, whether through live monitoring or analysis of previously collected information. Claude also cannot be used to recommend who should face investigation, arrest or criminal charges.
For applications affecting healthcare, finances, legal rights, employment or access to essential services, Anthropic continues to require qualified human oversight and disclosure that artificial intelligence contributed to consequential recommendations.
The revised rules also address artificial intelligence connected to equipment capable of taking physical actions. Such systems must have operators able to monitor their operation and intervene when necessary.
Equipment must additionally be capable of maintaining a safe condition if its connection to Claude is interrupted.
Anthropic said many provisions clarify restrictions already applied rather than establish entirely new prohibitions. The changes reflect increasingly autonomous capabilities and misuse patterns identified through its monitoring activities.
The company also clarified geographical access restrictions affecting organisations controlled by entities based in unsupported regions, alongside existing prohibitions on access from those territories.
Topics
Technology