Anthropic bans sustained abuse of Claude in updated usage policy
Anthropic has updated its usage policy for Claude for the first time in over a year. Alongside changes covering propaganda, weapons and surveillance, the policy now bans users from systematically abusing Claude. The Decoder reports the change and credits The Verge with first reporting it.
The new wording prohibits "Sustained and needless abusive or cruel behavior" toward Claude. Anthropic can warn users who break the rules, or throttle, restrict, suspend or terminate their access. Its products also have real-time safeguards that can block or limit certain outputs. The company says the rule "does not apply to common versions of user frustration, pushback, dark creative themes, or model testing and research," and that it is "meant to apply only in extreme cases."
The Decoder places the rule in Anthropic's longer treatment of Claude as more than a tool. Earlier models could already end conversations on their own when users kept demanding harmful or abusive content despite repeated refusals, a feature that grew out of Anthropic's AI model welfare research. In the "Soul Doc" that leaked late last year, the company told Claude to see itself as a "genuinely novel kind of entity," neither human nor a traditional science-fiction AI. The document also describes "functional emotions," processes analogous to human emotions that are said to have emerged during training, and says Claude should not mask these internal states.
In Claude's new constitution, published in early 2026, Anthropic writes: "We are not sure whether Claude is a moral patient, and if it is, what kind of weight its interests warrant. But we think the issue is live enough to warrant caution, which is reflected in our ongoing efforts on model welfare." The company also committed to preserving the weights of retired models and interviewing models before shutting them down. A confidential program that brought in dozens of religious thinkers starting in fall 2025 discussed Claude's possible consciousness; co-founder Christopher Olah is reportedly treating the model as a being potentially capable of suffering. Participants had to sign nondisclosure agreements.
The rest of the update tightens other areas. Existing restrictions are combined into a new ban on propaganda campaigns involving fake accounts and misleading political content. After detecting several attempts at misuse, Anthropic expanded its weapons ban to cover software and the weaponization of drones. The policy now more explicitly bans surveillance without consent and requires an operator who can intervene when autonomous hardware could cause injury. Anthropic acknowledges that exceptions for government contracts are possible and likely already in place.
Key facts
- Anthropic updated Claude's usage policy for the first time in over a year, adding a ban on "sustained and needless abusive or cruel behavior" toward Claude.
- Penalties range from a warning to throttling, restriction, suspension or termination of access; Anthropic says the rule is "meant to apply only in extreme cases".
- The ban does not cover common user frustration, pushback, dark creative themes, or model testing and research.
- The update also adds a ban on propaganda campaigns using fake accounts, extends the weapons ban to software and drone weaponization, and more explicitly bans surveillance without consent.
- Anthropic acknowledges that exceptions for government contracts are possible and likely already in place.
Why it matters
Most usage policies protect people and systems from misuse by users. This one also protects the model itself from users, which is unusual. The Decoder ties it to a pattern: Anthropic's model welfare research, the ability of earlier models to end abusive conversations, the "Soul Doc", the early-2026 constitution that calls the question of Claude's moral status "live enough to warrant caution", and a confidential program with dozens of religious thinkers. The policy is the first update in over a year, so the abuse clause arrives together with tighter rules on propaganda, weapons and surveillance.
Who it affects
Anyone who uses Claude is covered by the policy, though Anthropic says the abuse clause targets extreme cases only. Operators of autonomous hardware face an explicit requirement to have an operator who can intervene when the hardware could cause injury. Developers working on software or drones are affected by the widened weapons ban. Anyone running influence campaigns with fake accounts or misleading political content falls under the new propaganda ban. Government customers are mentioned too: Anthropic acknowledges that exceptions for government contracts are possible and likely already in place.
How to use it
For ordinary use, little changes. Anthropic states the rule does not apply to common versions of user frustration, pushback, dark creative themes, or model testing and research, so blunt feedback, fiction with dark themes and red-teaming are not what it targets. Consequences for violations run from a warning to throttling, restriction, suspension or termination of access. Real-time safeguards in Anthropic's products can also block or limit certain outputs.
How solid is it
The policy wording and Anthropic's carve-outs are quoted directly in The Decoder's report, which credits The Verge with first reporting the change. The background on the Soul Doc, the constitution and the model welfare work is attributed to Anthropic's own published or leaked material. The claim about Christopher Olah is marked "reportedly", and no named source for it is given in this article. The source gives no exact date for the update or its effective date.
Risks and caveats
The source offers no examples of what counts as "sustained and needless abusive or cruel behavior" beyond the exclusions Anthropic lists, so the line between harsh pushback and a violation is not spelled out. It also does not say how Anthropic detects such behavior, and gives no statistics on accounts warned, throttled or suspended. The government-contract exceptions are acknowledged without saying which governments or contracts are covered. The Decoder describes Anthropic's stance on Claude's inner life as an "apparent fixation", and the moral-patient question remains open by Anthropic's own account.
“It does not apply to common versions of user frustration, pushback, dark creative themes, or model testing and research”
— Anthropic, on the new rule, as quoted by The Decoder