Sensitive data is judged by meaning and intent
Data protection is AI Security Runtime™ running on the data path, beneath the workspace. Every prompt, file, and response crosses the same four functions on the way to the model.
| READS MEANING | The Semantic Security Engine judges each interaction in context, and sensitive-data detection improves as decisions accumulate in the Security Context Graph™ |
| REDACTS IN PLACE | The sensitive span is removed inline while work continues |
| COVERS BOTH WAYS | Prompts and responses pass the same judgment across the Secure Enterprise AI Workspace, gateway, API, agents, and Endpoint Security for AI tools you do not own |
| PROVES EACH DECISION | Every enforcement decision creates a metadata record of what ran, what was enforced, and why, without duplicating prompt or response bodies |
Common questions
What does it catch?
Credentials and access keys, payment data, customer and employee records, regulated content, and the sensitive data you define yourself. Detection is semantic: the engine weighs what the data is, who is sending it, where it is headed, and what the interaction intends, and it grows more accurate as your organization uses AI and connects its data, drawing on the Security Context Graph™.
What happens when a detection fires?
Policy decides the outcome per detection: allow and record, redact in place, hold for human review, or block. Redaction removes the sensitive span and lets the interaction continue, and hold routes ambiguous cases to a reviewer instead of stopping everyone. Deterministic rules stay deterministic: a class you designate always-block is blocked regardless of context; semantic judgment adds context on top, it never overrides a hard rule.
Do you store our prompts?
Analysis happens in the decision path. What is stored and exported is the decision and its context, sealed and metadata-only: surface, detection, policy, action. Raw prompts and data are not retained as evidence. The deliberate exception is hold for human review: a held item is visible to the reviewers you designate while it waits, and once decided, what remains in the trail is the decision, not the content.
The policy behind every decision
Decisions come from policy you control. Detection patterns carry severities, actions, and framework mappings, written once and enforced on every surface the runtime governs.
