Please report security issues privately via GitHub's "Report a vulnerability" (Security Advisories) on this repository. Do not open a public issue. We aim to reply within 5 working days.

• Providers can only choose among closed options that the library built from your schema. Any other answer makes the result unavailable, and your executor is never called.
• Numbers and dates are parsed by code. Entity ids come only from your resolver.
• execute() re-validates filters strictly, and runs authorize again, on every call.
• Jev keys are never logged, cached across requests, or included in results or errors.
• User text goes to the provider as state, separate from the instructions compiled from your schema.
• The core library sends no telemetry.

• That the model understands every request. An allowed filter can still have the wrong meaning. Show the interpretation chips to users.
• That your executor, resolver, or authorize are correct. Your executor must apply tenant/user scope from the trusted session as an outer AND, and use least-privilege, read-only credentials.
• Data handling by the model provider. The search text and your filter schema (field names, labels, descriptions, and allowed values) are sent to the configured Jev endpoint. Your records, credentials, and query code never are. If your category names or descriptions are sensitive, treat them like the search text. Review TypeSafe's terms before sending sensitive text.