The free floor masks the basics. Data Shield masks the rest.
Every plan masks structural fields before anything reaches an external AI provider. AI Data Shield adds name, organization and financial-figure masking, coverage for tool-call arguments, and a receipt for every message.
What leaves, and what doesn't
The same three hops, masking more by the time a message reaches the provider.
An admin can also turn on a stricter setting that withholds row-level detail from the model entirely, rather than masking it.
Masking that covers tool calls, not just chat
AWS Bedrock Guardrails' sensitive-information filter, by its own documentation, "evaluates text content only" and does not check the arguments, results, or definitions of a tool call. LinearPilot's masking runs through tool-call arguments the same way it runs through chat text, and restores the real values in the response.
A receipt for every message
Every masked message keeps a record of what was substituted: which categories, how many instances, and under which policy. An auditor can ask what left in a specific message and get an answer instead of a guess.
Industry packs
Each pack sets the posture that industry needs, and turns on any identifier categories that apply.
Law firms: matter names
Accounting: EINs and bank routing numbers
Insurance: policy-number templates
Healthcare: patient names and the Medicare Beneficiary Identifier
Real estate: client and property names, plus Korean road-name addresses
What it costs
AI Data Shield is a plan add-on, priced in the plans console. Your account's current rate is on the pricing page.
See pricing