Back to AI/ML Overview
🔐 Knowledge Site · Authentication and Authorization · Tokens, Policy, Workload Identity, Agents

Auth EngineeringA valid token is not a permission.

Authentication proves who is calling; authorization decides what that caller may do to which record, at this moment. This site starts from one broker reading one insurance quote and climbs, one idea at a time, through sessions, authenticators, API keys, tokens, OAuth and OIDC, access-control models, policy engines, tenants, certificates, workload identity, Kafka, cloud IAM and private environments, up to that call models, and retrieval while acting for people.

🎯The problem

A broker logs in, an agent gathers evidence, and an underwriter records a decision. A successful login, a valid JWT, an connection or an x-app-id header each looks like trust, yet none of them says whether this caller may read this quote or write this decision. Auth engineering puts an explicit, audited, resource-specific decision at every hop, and keeps the agent's own identity apart from the authority a person delegated to it.

🔑The question every page asks
Who is acting, on whose behalf, and may they do this to this record now? One synthetic underwriting workflow and one coding-agent pipeline, with negative tests proving no cross-tenant read, no restricted and no underwriting write by an agent.

🗺️What is inside

From one login to tokens

Identity primitives, people and proofing, sessions and browser security, passkeys and MFA, API keys, JWT anatomy, OAuth, OIDC and federation, per current standards.

Authorization

Decisions from first principles, , ABAC and ReBAC, PEP, PDP and PIP, policy as code with OPA, Cedar and OpenFGA, , governance and privileged access.

Workloads and the enterprise

TLS and PKI, SPIFFE workload identity, gateways and zero trust, Kafka ACLs, cloud IAM without secrets, webhooks and private controlled environments.

Agents and operations

Model providers, , retrieval ACLs, agents as delegates, A2A trust, coding agents and CI, audit, threat testing, rotation and revocation, framework profiles and playbooks.

🖱️How to read it

  • New to identity: read in order. Each chapter starts in plain English with one tiny example.
  • Architects: jump to enforcement, tenants, workload identity or agents as delegates. The rigorous part sits in folded blocks.
  • Fourteen interactive labs, among them a four-hop request explorer, a token validator, a policy decision lab and a delegated-agent workflow.
  • The example is synthetic and its tokens use obviously fake keys. Results show how the mechanism works; production code should use maintained libraries.