Anthropic Announces New Policies Prohibiting Abuse Against Its Claude AI Model

For the first time in over a year, artificial intelligence company Anthropic has updated its usage policy to ban continuous and unnecessary mistreatment and cruelty toward the Claude model. The new rules introduce stricter restrictions in high-risk areas such as election interference, weapon development, surveillance activities, and deceptive campaigns.

New Model Welfare Criteria for Claude

Anthropic has introduced a new rule aimed at preventing extreme and unnecessarily cruel behavior exhibited by users toward the artificial intelligence model. The company stated that this policy will only be applied in extreme cases where users repeatedly act cruelly toward their models without any clear purpose.

It was stated that this regulation does not cover ordinary user frustrations, criticisms, creative dark themes, or model testing processes. It was also noted that the conversation termination method will be maintained as the primary enforcement mechanism for the process.

Election Security and Deceptive Campaigns

With the new policy update, previously scattered restrictions were gathered under a common ban against deceptive commercial and political campaigns. Efforts to conceal the identities behind messages against AI-backed propaganda have been limited.

In the elections section, all attempts to spread misinformation about candidates or voting processes, impersonate candidates or officials, or obstruct voter turnout have been strictly prohibited.

Weapon Development and Surveillance Activities

The company further expanded its existing ban on the use of the Claude model in weapon development processes. Accordingly, software and components enabling weapons to operate, as well as the arming of autonomous vehicles such as drones, have been added to the scope of the ban.

In the field of surveillance, non-consensual human tracking and tracking conducted through the analysis of real-time or historical data were banned. It was clarified that the Claude model cannot be used in law enforcement processes to suggest who should be investigated, arrested, or charged.

Autonomous Hardware and Sectoral Approaches

Anthropic has introduced the rule that if artificial intelligence models are connected to hardware capable of performing autonomous physical actions that could cause injury, a qualified operator must observe and be able to stop the equipment.

On the other hand, the approaches of other players in the industry regarding model welfare vary. Mustafa Suleyman from the Microsoft AI team argued that AIs should not have rights or legal personhood, while debates on consciousness science continue.