Security Protection Program
Utilizing the SDK'sGuardrailsThe mechanism establishes three levels of protection:
- Ex-ante Inspection Layer::
- utilization
@input_guardrailDecorator creates preflight function - The data is transmitted through specialized intelligences (e.g.
guardrail_agent) Analyzing Input Sensitivity - define
PydanticModel specification check output format
- utilization
- process control layer::
- set up
tripwire_triggeredThreshold Trigger Interrupt - Returns for high-risk operations (e.g. math calculations)
403status code
- set up
- ex post facto audit level::
- combining
TracingFunctionality logging full interaction logs - Tagging of triggered guardrail events for storage
- combining
Example of implementation: When creating a math homework testing guard, it is recommended to set thereasoningfield describes the reason for the block, both blocking the request and enhancing the user experience.
This answer comes from the articleOpenAI Agents SDK: A Python Framework for Building Multi-Intelligence Collaborative WorkflowsThe
May not be reproduced without permission:AI productivity tools " How to address security issues when intelligences handle sensitive requests?
English 
简体中文 
日本語 
Deutsch 
Português do Brasil