The rapid integration of AI agents into applications necessitates robust access control to prevent unauthorized use and ensure compliance, which is being addressed by PydanticAI in conjunction with Permit.io. This framework simplifies the development of AI agents by providing structured input validation, response management, and security enforcement, enabling fine-grained access control through a Four-Perimeter Framework. This approach includes prompt filtering to block unauthorized queries, data protection to prevent unauthorized data exposure, secure external access to control interactions with third-party systems, and response enforcement to ensure compliance with regulations. The implementation utilizes attribute-based policies via Permit.io's no-code UI and PydanticAI for structured validation, ensuring that AI systems are both efficient and secure. This model serves as a foundational strategy, recognizing that as AI security evolves, it will require continuous refinement and adaptation to address emerging challenges and opportunities in real-world deployments.