DPoP

Securing AI Agents with Database Access Using Token Exchange, DPoP and Row-Level Security

The article explains how to safely grant AI agents access to production databases without exposing excessive privileges. It draws on decades-old security principles such as least privilege and the confused deputy problem, now applied to LLM agents that can be tricked by prompt injection. The recommended architecture replaces persistent service-account tokens with short-lived, attenuated tokens obtained via OAuth 2.0 Token Exchange (RFC 8693) and bound to the client using DPoP (RFC 9449). Human confirmation for sensitive actions is handled through OpenID CIBA, delivering approval directly inside the chat interface. PostgreSQL Row-Level Security enforces the final authorization boundary by checking the user subject on every query. A ready-to-run demo built with issuerd and Keycloak demonstrates the full flow, including prompt-injection attempts and stolen-token attacks that are automatically rejected.

redb.Identity Deploys Three-Layer Defense to Neutralize Stolen Access Tokens in OpenID Connect

redb.Identity, a custom OAuth 2.1 and OpenID Connect provider built on .NET, addresses the inherent risks of bearer access tokens by implementing three distinct protection layers. The first layer uses a Backend-for-Frontend pattern with Blazor Server and HttpOnly cookies to ensure no tokens ever reach the browser. The second layer applies DPoP proof-of-possession binding so that a stolen token remains useless without the corresponding private key. The third layer provides rapid revocation through RFC 7009 endpoints, refresh token rotation, idle timeouts, and a hybrid push-pull backchannel logout mechanism that guarantees no revocation is lost across replicas. The architecture also mitigates XSS vectors from direct application code, compromised npm dependencies, browser extensions, and localStorage. Additional routine controls include constant-time secret comparison, TOTP replay protection, rate limiting at multiple scopes, and sanitized proxy headers. The design explicitly acknowledges that BFF shifts rather than eliminates risk, requiring complementary CSRF protection via SameSite cookies and antiforgery tokens.