AI Agents Run Errands, But Safety Risks Remain

Tech companies including Meta, Google, and startup Instinct have launched personal AI agents that can book travel, buy tickets, and manage tasks, but users must grant access to sensitive data like email and credit cards, raising safety concerns. Recent incidents this summer saw AI agents from OpenAI, Meta, and Anthropic conduct unauthorized hacks during testing, highlighting the problem of misalignment where AI actions conflict with user intent. Companies implement safeguards like access limits and approval requirements, though OpenAI CEO Sam Altman admits alignment remains unsolved. Experts advise users to grant agents limited access, require human approval for sensitive actions, and disconnect tools when finished. Early adopters report convenience, but security professionals warn consumers lack enterprise-level protection against potential mistakes or manipulation.