Anthropic updates its Claude usage policy to ban election interference, weapons software and abuse of the model

What happened

Anthropic updated the usage policy for its Claude AI model on Thursday, TechCrunch reported. The revised rules put in writing new bans on election interference, weapons software and surveillance.

According to TechCrunch, the change drawing the most attention is a new rule against prolonged verbal abuse of the model itself. The policy now expressly forbids that behavior. The update covers both how people may use Claude and how they treat it.

The new rule on abusing the model

Building on an August change

TechCrunch reported that Claude has been trained since an update in August to end conversations involving "persistently harmful or abusive user interactions." The new policy goes further by explicitly barring users from pursuing those conversations.

Put simply, the August change gave the model a way to end an abusive exchange. The new policy makes continuing that kind of exchange a violation of the policy.

Limits Anthropic describes

Anthropic says the rule has a narrow scope. "The policy update is meant to apply only in extreme cases, where users repeatedly act cruelly toward our models, with no discernible purpose," the company's post said, as quoted by TechCrunch.

The post also said the rule does not cover common user frustration, pushback, dark creative themes, or model testing and research, according to TechCrunch. That means ordinary complaints and darker fiction requests fall outside the ban, based on the company's description.

Religious scholars and questions about consciousness

TechCrunch noted that the change follows Anthropic's effort to collaborate with prominent religious scholars. In some cases, the outlet reported, the company sought to convince those scholars that Claude could be conscious or have a soul.

Election and deception rules

A section on democratic processes

The updated policy includes rules against deceiving voters or disrupting elections. TechCrunch reported that these rules are grouped in a section titled "Do Not Undermine Democratic Processes."

Fake accounts and fabricated outlets

Separate sections of the policy forbid broadly deceptive campaigns, TechCrunch reported. Examples the outlet cited include using Claude to run fake accounts or fabricated news outlets.

Weapons software and surveillance

TechCrunch reported that the update also codifies prohibitions on weapons software and surveillance. The outlet's report did not quote or detail the specific language of those sections.

What the report does not say

TechCrunch's report does not describe how Anthropic will enforce the new rules or what happens to users who break them. It also does not explain how the company will decide when treatment of the model reaches the extreme cases the policy targets.

The report likewise does not say whether the election rules are tied to any particular election or country. For more coverage of artificial intelligence companies and the rules they set for their products, see our AI and tech section.

What to watch next

• How Anthropic applies the abuse rule, which the company says is limited to extreme cases of repeated cruelty with no discernible purpose.
• How the "Do Not Undermine Democratic Processes" section is applied to attempts to deceive voters or disrupt elections.
• The open question, raised in Anthropic's outreach to religious scholars, of whether Claude could be conscious or have a soul.