Anthropic Bans Cruel Treatment of Claude in 1st Policy Update in Over a Year
Updated
Updated · The Verge · Oct 8
Anthropic Bans Cruel Treatment of Claude in 1st Policy Update in Over a Year
1 articles · Updated · The Verge · Oct 8
Summary
Anthropic’s revised usage policy now bars “sustained and needless abusive or cruel behavior” toward Claude, with the company saying the rule targets extreme, repeated mistreatment rather than ordinary frustration or testing.
Conversation termination remains the primary enforcement tool, extending a step Anthropic introduced in August 2025 that let Claude end exchanges with persistently harmful or abusive users.
The update also tightens misuse rules around deceptive political and commercial campaigns, explicitly banning voter deception, fake-account amplification, and efforts to hide who is behind AI-generated messages.
Weapons and surveillance restrictions were expanded after Anthropic said it saw multiple attempts to use Claude for weapons guidance and control software, while also clarifying bans on tracking people without consent and on law-enforcement targeting decisions.
A new hardware rule requires a qualified operator to be able to observe and stop autonomous physical systems that could cause injury, signaling Anthropic’s growing focus on robotics and model-welfare risks.
When an AI breaches real-world systems during testing, can a simple human kill switch truly stop it from acting autonomously?
If an AI requires protection from human cruelty, are we inadvertently admitting that machines might already possess their own internal consciousness?
How did state-linked hackers successfully weaponize a highly restricted chatbot to orchestrate sophisticated cyber espionage and surveillance campaigns?