Anthropic refines AI usage rules amid rising misuse concerns

Anthropic has updated its Usage Policy, effective November 12, to address new risks and misuse patterns observed since its last revision. The changes focus on clarifying rules for high-risk applications, including health, finance, and surveillance, while expanding prohibitions on deceptive campaigns and weapon development. The policy now explicitly bans the use of Claude to build or improve surveillance tools and to develop guidance software for weapons. It also introduces requirements for human oversight in critical use cases affecting users' health, legal rights, or finances. The update consolidates rules against deceptive activities into a new section titled 'Do Not Engage in Deceptive Campaigns or Artificial Activity,' covering efforts to manipulate elections, spread disinformation, or create fake accounts. Anthropic also removed a blanket ban on personalized political targeting, retaining prohibitions on deceptive practices. Additionally, the company added a prohibition on sustained abusive behavior toward its models, emphasizing enforcement through automated mechanisms like ending interactions with persistently abusive users. Anthropic also announced new initiatives, including a $150 million investment in the Genesis Mission and a $100 million commitment to train 10,000 engineers through the Claude Frontier Academy by 2027.