CoinWorld News: Anthropic has introduced a brand-new security architecture for its team collaboration feature, Claude Tag, completely doing away with account sharing and directly assigning each AI its own dedicated, independent account. This new architecture replaces the old model where AI acted on behalf of employee accounts, ensuring that when posting on Slack, submitting code on GitHub, or querying a data warehouse, the AI runs using its exclusive account. Enterprise administrators can centrally configure the tools and code repositories that AI is allowed to access, and finely adjust access levels across different channels to ensure sensitive conversations don’t get leaked. In the future, the system will also support just-in-time authorization to reduce permission abuse to the greatest extent.