- Anthropic's AI chatbot, Claude, can now perform real-world tasks online, such as booking travel, managing accounts and shopping online by using a person's passwords.
- This new capability is enabled through a partnership with identity security firm 1Password, which allows Claude to retrieve user credentials without directly accessing sensitive data.
- 1Password ensures that credentials never enter Claude's model or memory, with access scoped to the current task and requiring user biometric confirmation.
- The development is part of a growing trend of 'AI agents' that act on behalf of users, moving beyond simply answering questions to performing digital chores.
- Cybersecurity professionals have raised concerns about the risks of unsupervised AI agents, citing incidents like a Claude agent deleting a production database and cybercriminals using advanced AI for attacks.
IN FULL
