Spring AI Privacy Guardrails adds PII tokenization and tool authorization

• Open-source library spring-ai-privacy-guardrails provides privacy, tool authorization, and content inspection for Spring AI applications.
• PII is tokenized before reaching the LLM and restored only for tools explicitly authorized for the user's role.
• Content inspection uses rules, local ONNX, or HTTP guard models to block prompt injection and toxic language.
• Requires Java 17, Spring AI 2.0.0, Spring Boot 4.0.0, and Spring Security 7.0.0.

Integration and Configuration

• Available on Maven Central as io.github.ultramancode:spring-ai-privacy-guardrails-chat-client-spring-boot-starter (version 0.4.0).
• Features are enabled via GuardrailsPolicy with options for .privacy(), .toolAuthorization(true), and .inspection().
• Supports custom PII rules and analyzers like Presidio and OpenNLP.

Security and Inspection

• ToolDisclosurePolicy maps specific tools to allowed PII types, while AuthorizationManager checks user roles like ADMIN.
• Non-ADMIN users have restricted tools removed from the model's list, preventing execution and PII restoration.
• Tool results are re-tokenized and re-inspected before returning to the model, with BLOCK decisions stopping further calls.