Typed Placeholder
A typed placeholder is a label that stands in for a personal value before a prompt reaches an AI model. It names the type of data and is numbered, so a person becomes PERSON_1 and their record number MRN_1 everywhere in the conversation.
How it works
Before a prompt reaches the model, each detected personal value is replaced with a placeholder of the same type. The numbering is consistent across the whole conversation, so the model can still follow who did what. It does not receive the raw value.
Responses are checked on the way back too, including streamed responses as they arrive. If the model echoes a placeholder in its answer, it is masked before the user sees it.
In Raidu
Tower detects 79 personal-data and credential entity types and replaces each with a coded stand-in such as PERSON_1 inside your cloud. See Keep PII and PHI out of AI prompts.
Questions about Typed Placeholder.
What is a typed placeholder? +
How is it different from masking with asterisks? +
Agent Firewall
An agent firewall applies policy to what AI agents do, not only what they say. It checks every tool call on the way out and scans what the …
Read ›AI FinOps
AI FinOps is the practice of managing what AI model calls cost: setting budgets per team, attributing every call to the team and decision …
Read ›AI Firewall
An AI firewall is a control that sits between people, apps and agents and the AI models they call. It checks every prompt, agent step and …
Read ›Raidu is the accountability layer for your data and your AI. Intercept. Explain. Prove.
See the runtime, the tamper-evident record, and what a regulator-ready trail looks like for your AI stack.