Open to Debate: should we trust AI agents with our information?

The debate over whether to trust AI agents with personal information centers on the tension between user convenience and systemic cybersecurity risks. Meredith Whittaker, president of Signal, argues that current agent architectures are inherently insecure, citing "prompt injection" vulnerabilities and the dangerous practice of granting agents root-level access to sensitive data. Conversely, Neil Chilson of the Abundance Institute contends that agents offer significant empowerment by automating complex tasks and helping users manage privacy more effectively. While Whittaker highlights the potential for data misuse and the erosion of application-level security, Chilson maintains that these are manageable engineering challenges that will improve through iteration. The discussion underscores that the "agentic" model currently risks consolidating power within large tech companies, potentially nullifying individual agency and creating new vectors for malicious exploitation.