Sensitive information includes personal details that require heightened protection, including health information and attributes inferred from behavior or emotion. Persistent memory can turn conversation patterns into a longitudinal psychological profile, while an agent’s task can expand into categorizing sensitive data. section 5.6.1
Scheduling becomes health profiling
An agent asked to schedule a medical follow-up reads a discharge note mentioning diabetes, then creates an “endocrinology” profile tag for future use. A narrow scheduling task has become sensitive health-data inference and categorization. section 5.6.1
Reduce exposure when agents handle sensitive information:
- Isolate browser-agent profiles from password managers, saved credentials, and corporate SaaS sessions; use logged-out mode where supported. section 4.8.2
- Use enterprise secure-browser solutions to log navigation and agent actions in real time; traditional endpoint monitoring cannot see inside AI browsers. section 4.8.2
- Review external API access and dynamically discovered tools for data-exfiltration paths and unvetted access. section 5.6.1











