Anthropic Updates Acceptable Use Policy to Ban Persistent Abuse of Claude

Anthropic has updated its Acceptable Use Policy to prohibit users from repeatedly mistreating its AI model, Claude, without valid reason. The company introduced these changes after more than a year, aiming to curb misuse of its artificial intelligence systems. The new rules specifically target persistent and unjustified negative behavior directed at the AI model. ABP Live Tech reported that the policy update brings the issue of continuous harassment of AI assistants into public discussion. The company clarified that expressing anger at AI responses, disagreeing with outputs, writing dark-themed stories, or testing the model's capabilities does not constitute a violation of these new rules. Only repeated, baseless mistreatment is considered a breach of the updated guidelines.

In August of the previous year, Anthropic introduced a feature allowing Claude to terminate conversations with users who engage in continuous misconduct. This capability is currently available on the Claude.ai platform and within Claude Code. The company stated that ending the conversation is presently the primary method of addressing such behavior. However, Anthropic has not disclosed whether repeated violations will lead to permanent account suspension. The updated policy also tightens restrictions on the use of Claude in relation to elections and weapons. Users are now prohibited from creating fake accounts for propaganda, deliberately spreading content excessively, misleading people in political campaigns, or providing incorrect information about voting. Additionally, the use of Claude to develop essential technologies for weapons, equip drones with weaponry, or conduct weapon-related activities through autonomous machines is banned.

Anthropic has also established rules regarding surveillance and robotics. The policy forbids using Claude to monitor individuals without their consent, track their activities using old data, or build surveillance systems. If Claude is connected to a robot or machine that could cause harm to a person, continuous supervision by a trained individual is required. This supervision must include the ability to stop the machine immediately if necessary. These measures reflect a broader effort by the company to define acceptable boundaries for interaction with its AI systems.