The Know Your Agent Framework for Financial Services
Paysafe's Chief Architect Amar Akshat introduces a novel approach to securing AI agents in regulated industries through the "Know Your Agent" (KYA) framework. This framework addresses the fundamental challenge that traditional identity models—built around humans and system accounts with long-lived trust boundaries—cannot adequately secure autonomous AI agents. The KYA framework creates a registry where merchants' AI agents register and can be verified by other systems, establishing provenance and accountability. This approach recognizes that agents challenge the existing trust model where employees or system accounts are trusted for extended periods, instead requiring continuous verification and attestation for short-lived agentic credentials.
Human-in-the-Loop Governance and Compliance
The conversation explores how Paysafe balances AI automation with regulatory requirements by implementing a mandate-based approval system. Whenever financial transactions or data movement occurs, human attestation is required through a seamless user experience—such as face ID or passkey authentication. This approach maintains compliance within existing regulatory frameworks that don't yet recognize AI agents as autonomous actors. The system tracks the complete chain from intent mandate through cart mandate to execution mandate, ensuring accountability while keeping the user experience frictionless. Akshat emphasizes that compliance cannot be a checkbox exercise but must be embedded in the agent's code execution path using policy frameworks like Open Policy Agent.
Productivity Gains and Shadow AI Prevention
Rather than blocking AI tools due to security concerns—a common initial reaction in regulated industries—Paysafe chose to embrace generative AI with proper governance. Akshat explains that blocking AI simply eliminates visibility, as employees will use these tools regardless, creating shadow AI risks. By implementing proper frameworks and controls, Paysafe has achieved significant productivity improvements through generative code technologies while maintaining security and compliance. The approach includes rigorous testing against golden datasets, regression testing for model changes, and comprehensive logging of model provenance, training materials, and confidence scores to establish accountability and enable reproducibility.
