• Home
• Technology

Anthropic, the US artificial intelligence company behind Claude, has expanded its usage policy to prohibit a broader range of harmful activities, including election interference, large-scale disinformation, weapons-related software development and certain surveillance practices. The revised rules also explicitly prohibit repeatedly abusing the AI model without a discernible purpose.

The changes build on measures introduced in August, when Anthropic trained Claude to end conversations involving persistent harmful or abusive behavior. Under the updated policy, users are expressly barred from continuing interactions whose sole purpose is to subject the model to repeated abuse.

Anthropic distinguishes such conduct from ordinary frustration or criticism. The restrictions do not cover users expressing dissatisfaction with Claude's responses, testing its capabilities, conducting research on the model or exploring dark and controversial themes in creative work.

New Restrictions on Election Interference

The revised policy sets clearer limits on using Claude to undermine democratic processes. Prohibited activities include running large-scale disinformation campaigns, creating fake identities and accounts, and operating fictitious media organizations that publish fabricated news and information.

The rules also bar using the model to deceive voters or manipulate them through misleading tactics, as well as activities intended to disrupt elections or undermine their integrity. The measures are designed to restrict the use of AI in political manipulation and the spread of false information.

Anthropic has also introduced additional restrictions on using its models to develop weapons-related software and conduct surveillance activities that could violate the company's safeguards for high-risk AI applications.

Anthropic Examines the Nature of Advanced AI

The policy changes come as Anthropic intensifies its efforts to understand the advanced models it develops, including questions about whether sophisticated AI systems could possess some form of consciousness or subjective experience.

In recent months, the company has engaged in discussions and collaborations with researchers and scholars specializing in religion and philosophy. These efforts explore the nature of advanced AI and whether systems such as Claude could be considered entities with some form of consciousness or what might be described as a soul.

Tighter Controls as AI Adoption Expands

The updated policy reflects a broader shift across the AI industry toward defining not only what advanced models can do, but also which uses companies consider unacceptable.

These safeguards are becoming increasingly significant as AI adoption expands across business, software development, media, research and analysis. Concerns are also growing over the potential use of AI to spread political disinformation, influence voters and develop tools that could pose security risks.

The new rules also draw a clearer distinction between legitimate user behavior, including criticism and capability testing, and deliberate, repeated abuse without an identifiable purpose.

Anthropic's revisions underscore its move toward a stricter framework for Claude as AI systems become capable of carrying out increasingly complex tasks rather than simply generating text and answering questions.