Who Let the Agents Out? Securing AI Workflows the Right Way
summary
A cloud-native agent platform assigns short-lived, user-tied identities to agents using SPIRE/SPIFFE SVIDs, performs OAuth token exchange via Keycloak to mint tool-scoped tokens, and uses Vault or OpenB as a credential store to map tokens to legacy API keys. MCP wires agents to tools and a production MCP gateway on Envoy aggregates many MCP servers. Kajenti automates registration with Kubernetes operators, sidecars and webhooks; a Llama Stack prototype shows per-agent tooling scales to about 10 tools before gateway optimizations are needed.
AI-written summary · about
chapters (11)
- 0:00Agentic AI workflows
- 1:46Zero trust for agent security
- 3:12Context loss in agent-tool
- 4:35OAuth and token exchange
- 6:41Workload identity with SPIFFY and SPIRE
- 7:45Legacy tools and credentials
- 8:48Secure agent workflows
- 13:50MCP gateway for scalability
- 17:20Identity automation with Kajenti
- 18:47Future plans and workshop
- 20:13Q&A on prompt injection
related talks
27mKeeping Your Agents in Check: Layered Security for Agentic Platforms in Production
36mPutting AI Agents to Work: A New Era of Open Source Connectivity with MCP
01h 22mIdentity for AI Agents
25mIdentities and Authentication for Your Agents With Keycloak
44mMCP Security: Keep Your AI Agents from Spilling the Tea
31mArchitecting Secure Agentic Workflows on Kubernetes: A Financial Sector…
18mThe "Agentic Organisation": Governing Multi-Agent Developer Teams with Nutanix Agent Gateway and MCP
27mMCPIdentity: Keyless MCP Agent Authentication Patterns on Kubernetes With Keycloak…