Cedar Policy Language

What It Does

Cedar is a declarative authorization policy language created by Amazon and open-sourced under Apache 2.0. It lets you express fine-grained access control rules as human-readable permit/forbid statements evaluated against a principal-action-resource model. Amazon uses it internally to power AWS Verified Permissions and Amazon Verified Access. It has formal verification — policies can be mathematically proven to behave as intended.

Leash by StrongDM adopted Cedar as its policy substrate for AI agent governance, transpiling Cedar policies into eBPF rules, HTTP proxy configs, and MCP observer rules — demonstrating Cedar’s versatility beyond traditional IAM.

Key Features

Declarative permit/forbid model: Policies are human-readable statements with when conditions; forbid always wins over permit (deny-by-default)
Principal-Action-Resource structure: Maps naturally to authorization questions — “Can this entity do this action on this resource?”
Formal verification: Cedar includes tools to mathematically prove policy properties (e.g., “no policy permits admin deletion by non-admins”)
Entity-based evaluation: Policies reference typed entities with hierarchical relationships (groups, roles, resource trees)
Condition expressions: Rich when clause with attribute access, set operations, and hierarchical in checks
Multiple language SDKs: Rust (reference), Java, Go, TypeScript, Python, Wasm
Fast evaluation: Sub-millisecond policy evaluation; designed for inline authorization in hot paths
Schema validation: Optional schema enforcement ensures policies reference valid entity types and attributes

Use Cases

Application authorization: Replace scattered if/else permission checks with centralized, auditable Cedar policies
AI agent governance: Define what agents can access at file, network, process, and tool levels (as in Leash)
Multi-tenant SaaS: Tenant isolation policies expressed declaratively and verifiable via formal analysis
AWS Verified Permissions: Native integration for applications built on AWS
Policy-as-code pipelines: Cedar files version-controlled alongside application code, reviewed in PRs, tested in CI

Adoption Level Analysis

Small teams (<20 engineers): Usually overkill. Simple role-based checks in application code suffice unless you have complex multi-tenant authorization requirements.

Medium orgs (20-200 engineers): Good fit when authorization logic has grown beyond what’s manageable in application code. Centralizing policies in Cedar makes them auditable and testable.

Enterprise (200+ engineers): Strong fit. Formal verification, centralized policy management, and AWS-native integration align with enterprise compliance and governance requirements.

Alternatives

Alternative	Key Difference	Prefer when…
Open Policy Agent (OPA/Rego)	General-purpose policy engine; Rego is more powerful but harder to learn	You need policies beyond authorization (admission control, data filtering, compliance)
Casbin	Library-based, multiple model support (RBAC, ABAC, ACL)	You want a lightweight embedded library, not a standalone policy language
Cerbos	API-first policy engine, YAML-based policies	You want simpler policy syntax and a managed SaaS option

Evidence & Sources

Notes & Caveats

Amazon-controlled: While open-source (Apache 2.0), development is primarily driven by Amazon. Community contributions exist but governance is Amazon-led.
Younger than OPA: Cedar (open-sourced 2023) has less ecosystem maturity than OPA (2016). Fewer integrations, fewer community policies, less tooling.
Authorization-specific: Unlike OPA which handles arbitrary policy decisions, Cedar is purpose-built for authorization. This is a strength (simpler, verifiable) and a limitation (can’t do admission control, data filtering, etc.).
AWS gravity: Strongest integration story is with AWS services. Non-AWS usage is viable but has less vendor support.

Access Required