New to this? Read it in simple words
- Anthropic’s new rules for Claude start on November 12.
- Users may not be cruel to Claude again and again for no reason.
- Claude can end such chats, and that stays the main response.
- Election rules now focus on deceiving voters, and fake accounts are banned.
- Usage policy
- The list of things customers may not do with a company’s AI models.
- Model welfare
- Research into whether AI models could have experiences that matter, and how to treat them.
- Influence operation
- A hidden campaign, often with fake accounts, that tries to change what people believe.
Every company that sells AI models has a usage policy: the list of things customers may not do with them. On October 8, Anthropic published a new version of its policy for Claude, which takes effect on November 12, 2026.
A rule about cruelty
The new policy says users may not “engage in sustained and needless abusive or cruel behavior toward our models”. The version in force since September 2025 had no such line.
Anthropic says the rule applies only in extreme cases, when users repeatedly act cruelly with no clear purpose. It says frustration, pushback, dark creative themes and model testing are not covered.
Claude’s ability to end such chats “will remain the primary enforcement mechanism,” Anthropic writes. The Verge reports that Anthropic did not say whether abusive users could also lose their accounts.
The main changes in Anthropic’s usage policy of October 8, compared with the version in force since September 2025.
Why a model gets a rule
Anthropic says only that the rule fits with Claude’s existing ability to end rare conversations with persistently abusive users. The Verge and The Register link it to Anthropic’s research on model welfare, which led to that feature in August 2025.
The Register points out that there is no established evidence that Claude feels distress. It also asks how Anthropic will tell real frustration apart from cruelty.
New election and campaign rules
The election section now focuses on deceiving voters and disrupting elections: false information about candidates or voting, pretending to be officials, and keeping people from voting. Election interference was already banned before.
Anthropic removed a blanket ban on personalised vote and campaign targeting. It says that ban also blocked useful work, such as nonprofits translating voter information. Deceptive targeting stays banned under other sections.
A new section bans deceptive campaigns and artificial activity, such as fake accounts, fake news sites and tools for influence operations. The policy also spells out rules for weapons software, surveillance, and Claude connected to machines that can hurt people.
Sources
Every fact in this story comes from the sources below. Open them to check our work.
- 1
- 2
- 3Research · October 8, 2026Anthropic changes usage policy to ban model abuse and election interference TechCrunch
- 4
- 5
The rules are quoted from Anthropic’s policy and compared with the 2025 version. The link to model welfare is the outlets’ reading, not Anthropic’s stated reason. Silicon AI News is written with Anthropic’s Claude models.
The main news is a new ban on “sustained and needless” cruelty towards Claude, a restriction that critics question, softened by looser campaign-targeting rules. How we rate the mood