AI coding agents are part of the developer toolchain. Tools like Kiro and Claude Code generate features, tests, and code refactors from natural-language prompts. A single agent can open dozens of pull requests (PRs) across your repositories in an afternoon. That productivity comes with a trade-off: agents optimize for task completion at machine speed with no understanding of your organization’s risk.

Through protocols like the Model Context Protocol (MCP), agents also reach beyond the integrated development environment (IDE) to call APIs, query databases, and modify infrastructure and even entire environments, expanding the scope of resources your application security team defends.

This post lays out an application security (AppSec) control framework for AI coding agents. Two pillars organize the framework: author-time controls shape what the agent produces in the IDE; build-time controls verify and gate what reaches production. Your existing secure software development lifecycle (SDLC) controls still apply and are critical to a defense-in-depth security strategy. The framework shows where to layer additional guardrails so AppSec scales with agent-driven development. The framework is tool-agnostic and cloud-agnostic. Throughout, we use AWS services—Kiro in the IDE and AWS CodePipeline in the build—as a running example that you can adapt to your own toolchain.

Each of the following risks includes a treatment summary. The control framework section later in this post provides implementation details. The risks are ordered by severity with the highest impact risks first.

Agents read untrusted content, such as issue descriptions, web pages, MCP responses, and README files in third-party packages. Text from outside parties can redirect the agent to disclose secrets, open unauthorized PRs, or invoke tools without user consent. This risk, known as prompt injection, is the top risk in the OWASP Top 10 for LLM Applications. Any agent that reads content from outside parties is exposed, with or without MCP, so connecting tools widens the scope of impact.

Treatment: Treat non-developer input as untrusted. A large language model (LLM) can’t reliably separate instructions from data in a single context window, so architect for it: keep the agent that orchestrates trusted actions separate from the one exposed to untrusted content and grant the exposed agent only read-only, least-privilege access. Require human approval for irreversible actions. Use version-control steering files to prevent silent tampering.

Agents optimize for getting work done. Left unchecked, the code they generate can default to wildcard identity and access management policies, open security groups, and unencrypted storage, or embed sensitive values in code rather than referencing a secrets manager. Most coding agents now include safety mechanisms that make these outcomes less likely, but they remain imperfect, so you still need controls to account for the possibility.

Treatment: Security requirements in a steering document, plus policy-as-code scanning (Checkov, cfn-nag) in the IDE and pipeline. See Context as a security control.

Ungated code reaching production isn’t new, but AI agents amplify it. Machine-speed generation can propagate a flawed pattern across repositories before it’s identified.

Treatment: Branch protection rules requiring PR approval (a human-in-the-loop checkpoint), pre-commit hooks for security checks, and sandboxed agent runs that prevent direct pushes to protected branches. The right balance between human review and automated speed depends on the risk profile of the change. For many low-risk paths, automated checks alone might suffice, while higher-risk changes warrant a human checkpoint.

Agents don’t always distinguish current best practices from outdated patterns. They might recommend deprecated packages, reference library versions with new Common Vulnerabilities and Exposures (CVEs), and hallucinate package names that don’t exist, which can introduce risks of dependency confusion issues.

Treatment: Software Composition Analysis (SCA) in the pipeline (for example, Amazon Inspector code scanning or Dependabot) to flag vulnerable or unexpected dependencies. For additional control, resolve against a scoped registry like AWS CodeArtifact. Even without a fully curated registry, lockfile validation and allow-listing critical packages reduce exposure.

Through MCP and tool integrations, agents query databases, call APIs, and modify infrastructure. Without constraints on which tools and data an agent can reach, a single misconfigured integration provides unintended access to sensitive resources.

Treatment: Scope MCP servers to least-privilege tools and resources, enforce authn or authz on external connections, and audit tool invocations. The control point is the configuration file. Review it the same way you review AWS Identity and Access Management (IAM) policies.

Agents produce plausible-looking output. Code that compiles, passes linting, and looks reasonable can still be functionally wrong: misusing APIs, introducing subtle logic errors, or implementing security-sensitive operations incorrectly. Code that passes continuous integration (CI) but is wrong slips through review; code that fails to build is caught immediately.

Treatment: Layer deterministic verification (static application security testing (SAST), unit tests) with non-deterministic review (LLM-assisted screening against the specification). Neither catches everything alone.

Given a bug-fix prompt, an agent might also refactor surrounding code, disable an unreliable test, or reorganize imports. Unrequested changes introduce regressions and complicate review.

Treatment: A reviewed specification document that defines what must change and what must not, paired with a targeted review of the proposed changes. See Specifications as scope boundaries.

The preceding risks share a common thread: agents produce output faster than humans can review it, and they lack context to self-correct.

The following framework addresses this gap. It organizes controls into two pillars: author-time (pre-generation and post-generation of code) and build-time (in the pipeline, before code reaches production). Author-time controls shape what the agent produces. Build-time controls verify it. Neither is sufficient alone; together they reduce the volume and severity of issues that reach human reviewers.

Deterministic mitigations [D] produce the same result every time. Linters, SAST scanners, secrets detection, and policy-as-code match patterns against rules and define security invariants: no critical findings, no hardcoded secrets, and no wildcard IAM policies. Use them when the condition can be expressed as a rule. Organizations already have these and must continue enforcing them.

Non-deterministic mitigations [ND] use model judgment. They include steering documents, LLM-as-judge review, specification compliance checks, and scope-creep detection, and they evaluate intent rather than patterns. They catch novel issues that rules miss, but are probabilistic. Use them when evaluation requires context or reasoning across files. This is the new layer that AI-generated code demands, because agents produce code that can pass every deterministic check yet remain functionally wrong.

Human review [H] provides the final layer for the risk-based decisions neither tool type can make. Apply it where judgment is needed, not everywhere: routing every change to a person invites consent fatigue, where reviewers approve by reflex and the control loses its value. The default reflex is to route everything back to a human, but that isn’t always the right response—reserve human judgment for the decisions that genuinely need it.

The framework organizes controls into two pillars. Author-time controls (Pillar 1) shape what the agent produces in the IDE, before code is generated and just after. Build-time controls (Pillar 2) verify and gate that output in the pipeline, before it reaches production. The controls within each pillar are tagged deterministic [D], non-deterministic [ND], or human [H].

Author-time controls work inside the IDE, where the developer and agent still hold full context. They shape the prompt and the generated output before it ever reaches a pull request. The following controls apply at this stage.

Context as a security control [ND]

Control statement: Encode security invariants as natural-language constraints in a steering document that every developer environment consumes at session start. Addresses R002.
Many AI coding agent risks share one root cause: the agent lacks the security context an experienced developer carries implicitly. Your security team sets the policies, such as Amazon Simple Storage Service (Amazon S3) buckets require encryption, API gateways require mutual TLS, and credentials must come from AWS Secrets Manager. Developers don’t always have these requirements available when they’re building. They build what works, not what’s compliant. An AI agent amplifies this gap because it defaults to whatever pattern dominated its training data, with no awareness of your organization’s security posture.

A key mitigation is steering. Security teams write these invariants once as natural-language guidance in a steering document, then distribute them as shareable resources that developers consume in their IDE. The agent loads the file at session start and treats the contents as standing requirements:

• IAM policies must follow least-privilege principles; no wildcard Amazon Resource Names (ARNs).
• No hardcoded credentials in source code; use a secrets manager.
• Security groups must not allow unrestricted inbound access.

This shifts security left, before code generation begins. Steering biases generation toward secure defaults; it doesn’t guarantee them. Treat it as a strong default, paired with the following deterministic gates that block non-compliant code from merging. Security teams define the rules once and every developer environment inherits them automatically. Steering reduces the volume of issues that reach the pipeline, though it doesn’t replace downstream scanning.

How to write effective steering rules: Keep each rule specific and testable, scope it to a concrete risk class, keep the rule set concise so the agent can hold it in context, and iterate from the issues your scanners and reviewers surface.

Specifications as scope boundaries [ND]

Control statement: Require a reviewed specification before code generation begins. Define what must change and what must not. Addresses R007.

Spec-driven workflows turn vague prompts into reviewable specifications before code is generated. This creates a human checkpoint at the design phase, where security decisions are made:

• Requirements use testable notation that’s auditable before the agent writes a line of code. For example, the Easy Approach to Requirements Syntax (EARS): WHEN [condition] THE SYSTEM SHALL [behavior].
• Tasks are ordered in implementation steps, each mapped back to a requirement.

For bug fixes, specifications add a critical element: unchanged behavior documentation. This is an explicit list of behaviors that must continue working, giving the agent a written boundary against scope creep.

In this model, the specification becomes the primary artifact, code is a derivative of it. Human review effort concentrates on whether the specification solves the right problem with the right constraints, not on reading implementation diffs line by line.

Controlled tool access using MCP [D + ND]

Control statement: Scope each MCP server to the minimum set of tools the agent needs, and give it a dedicated, scoped-down credential rather than the developer’s own. Maintain an allowlist of reviewed MCP servers. Addresses R005.

MCP servers act as controlled gateways between the agent, the external tools, and data:

• Dependency management – An MCP server fronting your private package registry resolves dependencies against curated packages, not the public internet. This is a deterministic constraint on supply chain risk.
• Infrastructure tooling – Visibility into current resource configurations prevents templates that conflict with existing infrastructure.
• Scoped permissions – Each MCP server exposes a defined set of tools and resources. You choose exactly what the agent can access, supporting least-privilege at the integration layer. You supply that credential through the agent’s configuration (in Kiro, the env block of .kiro/settings/mcp.json). Avoid autoApprove: ["*"], which removes the human approval prompt on every tool call.

IDE code scanning [D]

Control statement: Run real-time static analysis in the IDE so security issues surface while the developer (and agent) still have full context. Addresses R002, R006.

Real-time diagnostics catch syntax errors, type mismatches, and configuration issues as the developer types. A malformed IAM policy is flagged before the agent builds further on it. Security-focused extensions (ESLint security plugins, Checkov, SAST) layer on top for immediate feedback while code is fresh in context.

Hooks: Automated guardrails at the point of action [D + ND]

Control statement: Attach deterministic checks to file-save events and non-deterministic verification to task-completion events. Addresses R002, R007.

• Shell command hooks [D] – Triggered on file save, these run a linter, formatter, or security scanner and produce the same result every time. They enforce hard rules.
• AI-powered hooks [ND] – Triggered on task completion. These prompt the agent to verify that the implementation matches the specification and check for any untested edge cases or files that were modified outside the task’s scope.

Build-time controls run in the pipeline after code is committed and before it reaches production. They verify and gate what the agent produced, catching what author-time controls did not. The following controls apply at this stage.

Layered security scanning [D]

Control statement: Run secrets detection, static analysis, dependency scanning, and infrastructure-as-code scanning in sequence. Fail the build on any critical finding. Addresses R002, R003, R004.

1. Secrets detection runs first because it’s cheapest and addresses a high-severity class of issue. It scans for hardcoded API keys, database connection strings, and credentials that AI agents might inadvertently include.
2. SAST scans source code for injection issues, insecure deserialization, and resource leaks. Custom rules can target AI-specific anti-patterns including overly broad exception handling, deprecated APIs, placeholder credentials, dynamic code execution through eval().
3. Software Composition Analysis (SCA) identifies known CVEs in dependencies. This is critical for AI-generated code, which might reference deprecated packages or hallucinate package names that open you to dependency confusion issues.
4. Infrastructure as code (IaC) scanning validates AWS CloudFormation, Terraform, and AWS Cloud Development Kit (AWS CDK) templates against security policies before deployment. Catches overly permissive IAM roles, unencrypted storage, and public-facing resources the agent created.

Each stage halts the pipeline on failure. Results export to a standard format (Static Analysis Results Interchange Format (SARIF)) for compliance auditing and flow downstream to human reviewers. The open source Automated Security Helper (ASH) bundles secrets, SAST, SCA, and IaC scanners behind one command that you can run locally and in AWS CodeBuild, emitting SARIF for the gates that follow.

Quality gates [D]

Control statement: Define pass/fail thresholds for each scan type. Block deployment on any critical or high-severity finding. Addresses R003.

Quality gates convert scan results into go/no-go decisions. Define thresholds for each severity: block on critical findings, require justification for highs, and track mediums. The gate is deterministic: if a threshold is breached, the pipeline stops. Exceptions require documented approval.

Differentiate blocking compared to advisory modes: hard failures on main, advisory on feature branches. Avoid gates becoming a friction that teams route around.

AI-assisted review [ND]

Control statement: Use an LLM reviewer to pre-screen every pull request for specification compliance, scope creep, and security anti-patterns before human review. Addresses R001, R006, R007.

• Specification compliance – Does the implementation match the requirements document?
• Scope verification – Were files modified outside the task’s stated scope?
• Security pattern review – Are there logic errors, misused APIs, or insecure patterns that pass SAST but violate intent?

This pre-screening focuses human reviewer attention on genuine risks rather than formatting or obvious issues. On AWS, AWS Security Agent (code review in preview at publication) checks pull requests against AWS-managed and custom security requirements. The reviewer screens and surfaces findings; the merge decision stays with a human.

A critical principle: the agent that wrote the code should not be the agent that reviews it. A separate session helps avoid self-confirmation bias, but a separate session alone doesn’t always avoid the generator’s blind spots, because two sessions of the same model can share them. Where practical, use a different model for review so the reviewer is less likely to inherit the same systematic weaknesses.

Human-in-the-loop review [ND + H]

Control statement: Require human approval on most pull requests, especially those touching security-sensitive or high-blast-radius code. Lower-risk changes might be eligible for agent-assisted or fully automated approval as tooling matures. Provide reviewers with scan results, LLM pre-screening output, and specification context to enable fast, informed decisions. Addresses R003.

Scale review depth to the risk of the change. Low-risk or boilerplate changes can take a lighter-touch review, while security-sensitive or novel-logic changes warrant mandatory deep review and a second reviewer.

Scanners catch known patterns but can’t judge whether code implements the intended business logic. Human review also serves to calibrate trust: teams build intuition about where agents excel (boilerplate, test writing) and where they’ve tended to struggle (novel business logic, security-sensitive operations), recognizing that this frontier shifts as models improve.

Place two approval gates: after security scans (reviewer focuses on correctness and business logic, with scan results as context) and before production deployment (final sign-off after integration testing). Treat human review as a secondary control, not a guarantee: reviewers are themselves non-deterministic and can miss issues, so human review layers on top of the deterministic gates rather than replacing them.

The framework is tool-agnostic, but AWS gives you building blocks for each pillar. The following services map directly to the controls described previously: Kiro for author-time guardrails, and CodeBuild and CodePipeline for build-time gates.

Kiro maps to Pillar 1: It puts the author-time controls in the IDE, where the developer and agent still share full context. Each feature in the following list implements one of those controls, configured in-repo under .kiro/ so the guardrails are version-controlled and shared across the team rather than set per developer.

• Steering documents – Markdown files in .kiro/steering/ load into the agent’s context at session start. Conditional inclusion using fileMatch (for example, ["**/*.tf"]) loads IaC-specific rules only when relevant.
• Specification-driven workflows – Three-phase specifications (requirements in EARS, design, and tasks) with review checkpoints. Bug-fix specifications capture unchanged behavior explicitly.
• Agent hooks – Triggered on file save, tool invocation, or task completion. Shell hooks run deterministic checks (linters, tests); Ask Kiro hooks run AI prompts for non-deterministic review. For example, a security pre-commit scanner hook can flag hardcoded credentials when the agent finishes a task.
• Property-based testing – Guided by a specification or hook, Kiro can generate property-based tests (for example, using the hypothesis library) that exercise hundreds of randomized inputs, probing edge cases a hand-written test suite would miss.
• MCP integrations – Connect Kiro to private package registries, internal docs, issue trackers, and infrastructure tooling, creating the controlled tool access pattern.

For enterprise environments, Kiro supports AWS IAM Identity Center for single sign-on and provides IP indemnity coverage for subscribers. Check the Kiro documentation for current Region availability.

CodeBuild runs each scanning tool (checking for secrets, SAST, SCA, and IaC) as a build action. A non-zero exit code fails the action, and the stage halts or rolls back according to its OnFailure setting. Findings export as SARIF to Amazon S3 for compliance, and CodePipeline action variables pass results to downstream approval actions.

• CodeBuild exit codes halt the pipeline on scan failures
• AWS Lambda invoke actions evaluate scan results against configurable thresholds and return pass/fail decisions
• Manual approval actions halt the pipeline, send Amazon Simple Notification Service (Amazon SNS) notifications, and link to review artifacts; decisions and reviewer identity are logged for audit

The following table consolidates the framework into a single view that includes each stage of the SDLC and the deterministic [D] and non-deterministic [ND] controls that apply there. Every stage carries both, a reminder that neither control type is sufficient on its own.

This post laid out a framework for adopting AI coding agents at machine speed without letting unreviewed risk reach production. It layers guardrails at two points:

• Author-time controls – Steering, specs, and scoped tools shape what the agent generates in the IDE.
• Build-time controls – Scanning, quality gates, and layered review verify it before it reaches production.

No single layer is enough: deterministic gates enforce hard rules, non-deterministic review catches what they miss, and human judgment is reserved for the decisions that need it. Together, they let AppSec scale with agent-driven development.

Where to start this week:

1. Start with steering and specs – Encode security requirements as steering and use specifications for new features. Highest impact, lowest effort. For a ready-made starting set, the open source Project CodeGuard (a Coalition for Secure AI project under OASIS Open, of which Amazon is a contributing member) publishes reusable steering rules for common risk classes—hardcoded credentials, IaC misconfiguration, supply chain, and MCP security—that you can adapt to your AWS environment.
2. Add deterministic pipeline gates – Integrate SAST, SCA, and secrets detection. Table-stakes regardless of AI usage.
3. Calibrate and iterate – Review what controls catch, adjust steering for recurring issues, and expand agent autonomy as trust builds.
4. Accountability – Developers remain accountable for the security of what they ship. AI agents accelerate development; they don’t transfer ownership.

More information:

• Kiro documentation
• AWS CodePipeline User Guide
• AWS Agentic AI Security Scoping Matrix

If you have feedback about this post, submit comments in the Comments section below.

If you’re building multi-agent AI systems, you need to prevent authorization scope from silently expanding as agents delegate tasks through multi-hop chains. Without proper controls, an agent can potentially act beyond what the originating user authorized, even when role-based access control (RBAC) policies are in place. The OWASP Top 10 for Agentic Applications classifies this risk as ASI03: Identity & Privilege Abuse.

This post shows you how to address the potential risk using a three-layer policy model built with Cedar, an open source authorization policy language, deployed on Amazon Web Services (AWS). The reference implementation uses OAuth 2.0 for authentication and Cedar for authorization. A trusted identity provider authenticates the originating user, then Cedar policies enforce authorization across three layers using verified token claims.

To enforce authorization at each hop in a multi-agent delegation chain, the reference implementation uses two AWS Lambda functions in sequence. A Model Context Protocol (MCP) adapter Lambda function normalizes inbound requests and cryptographically signs the originating user context. This prevents downstream tampering. A Cedar evaluator Lambda function evaluates three independent policy layers sequentially, halting on the first deny.

Table 1: Three-layer Cedar policy evaluation model

Cedar evaluates authorization but doesn’t establish identity. Before Cedar can evaluate context.originating_user.role or context.originating_user.mfa_verified, a trusted authentication layer must establish the user’s identity and produce verifiable claims. Steps 1–3 handle authentication; steps 4–10 handle authorization. The architecture shown in Figure 1 is described in the following lists:

1. The originating user authenticates with an OIDC-compliant identity provider (in this reference implementation, Amazon Cognito with TOTP multi-factor authentication (MFA)). The identity provider (IdP) issues a signed JSON Web Token (JWT) containing claims such as sub, role, amr (authentication methods), and session_id.
2. Amazon Cognito returns the signed JWT to the user.
3. The user passes the JWT and task request to the AI agent (MCP client). The agent carries the originating user context in the MCP _meta envelope.

4. The AI agent sends a Model Context Protocol (MCP) request to AWS WAF, which filters using CommonRuleSet, SQLiRuleSet, rate limiting, and body size constraints.
5. Amazon API Gateway (with Amazon Cognito authorizer) verifies the JWT signature against the user pool’s public keys and rejects invalid or expired tokens. Valid requests are forwarded to the MCP protocol adapter Lambda function, which applies Amazon Bedrock Guardrails content filtering.
6. The adapter extracts verified claims from the token and maps them to Cedar context attributes:
   1. JWT role claim : context.originating_user.role
   2. JWT amr includes MFA method: context.originating_user.mfa_verified = true
   3. JWT sub: context.originating_user.user_id
   4. JWT sid: context.originating_user.session_id
   5. JWT amr claim: context.originating_user.authentication_method

   The adapter then computes an HMAC-SHA256 signature over the user context (user_id, role, mfa_verified, authentication_method, and session_id in canonical order) using a key from AWS Secrets Manager.

7. The adapter constructs a signed request envelope and invokes the Cedar evaluator Lambda function.
8. The evaluator verifies the HMAC-SHA256 signature, retrieves L2 and L3 Cedar policies from Amazon Verified Permissions, and evaluates all three layers (L1, L2, and L3), halting on the first deny.
9. The evaluator emits an Open Cybersecurity Schema Framework (OCSF) 99001 audit event to Amazon CloudWatch Logs. Failed emissions fall back to an Amazon Simple Queue Service (Amazon SQS) dead-letter queue (DLQ).
10. Amazon CloudWatch dashboards and alarms monitor evaluation latency, deny rates, and DLQ depth. Alarm notifications route through Amazon Simple Notification Service (Amazon SNS).

Two mechanisms work together to protect identity across hops:

• Hash-based Message Authentication Code (HMAC-SHA256) ensures integrity and authenticity. Every downstream evaluator verifies this signature before trusting the context.
• OAuth 2.0 Token Exchange (RFC 8693) sets delegation scope using the on-behalf-of (OBO) pattern. When the orchestrator delegates to a downstream agent (data-bot), it exchanges the original token for a scoped OBO token that records who’s acting on behalf of whom and with what authority. The Cedar policies (detailed in Step 2: Three-layer policies) then check whether that scoped delegation is permitted and verify the originating user claims carried in the OBO token. Token exchange limits each downstream agent to only the delegated task’s scope instead of passing through the full original token. For enterprise deployments, use token exchange alongside HMAC. OAuth tracks who is acting on behalf of whom and with what scope. HMAC verifies that the context hasn’t been tampered with and came from a trusted source.

The following prerequisites are needed to deploy the reference implementation. Before you begin, clone the repository:

git clone https://github.com/aws-samples/sample-cedar-agentic-ai-authorization.git
cd sample-cedar-agentic-ai-authorization

Verify that you have the following:

• An AWS account with AWS Cloud Development Kit (AWS CDK) bootstrapped in the target AWS Region
• Python 3.12 or later, and Node.js 20 or later (for CDK command line interface (CLI))
• AWS CDK CLI (npm install -g aws-cdk)
• AWS credentials with scoped AWS Identity and Access Management (IAM) policy permissions for Lambda, API Gateway, Verified Permissions, Secrets Manager, Amazon Virtual Private Cloud (Amazon VPC), and CloudWatch

In this walkthrough, you define the Cedar entity schema and policies, deploy the infrastructure with AWS CDK, and integrate your identity provider.

In this step, you define a schema with two entity types (Agent and Tool) and two actions (invoke_tool and delegate_task) in the AgentAuthz namespace. Notice that there is no User entity. Instead, you carry the originating user’s identity in the evaluation context record, which is a structured data object passed alongside each authorization request.

{
  "AgentAuthz": {
    "entityTypes": {
      "Agent": {
        "shape": {
          "type": "Record",
          "attributes": {
            "trust_level": { "type": "Long", "required": true },
            "namespace": { "type": "String", "required": true },
            "registered_capabilities": {
              "type": "Set", "element": { "type": "String" }, "required": true
            },
            "lifecycle_stage": { "type": "String", "required": true }
          }
        }
      },
      "Tool": {
        "shape": {
          "type": "Record",
          "attributes": {
            "namespace": { "type": "String", "required": true },
            "risk_level": { "type": "String", "required": true }
          }
        }
      }
    },
    "actions": {
      "invoke_tool": {
        "appliesTo": { "principalTypes": ["Agent"], "resourceTypes": ["Tool"] }
      },
      "delegate_task": {
        "appliesTo": { "principalTypes": ["Agent"], "resourceTypes": ["Agent"] }
      }
    }
  }
}

This schema is deployed to an Amazon Verified Permissions policy store by the VerifiedPermissionsStack CDK stack. In the reference implementation, the schema file is located at cedar-entity-schema.json.

Agent topology and attributes

The following tables show the agents and tools registered in this reference implementation, along with the attributes the Cedar evaluator function retrieves from the entity store. The test scenarios that follow trace requests through this topology.

Table 2: Agent attributes

Table 3: Tool attributes

The orchestrator can delegate to both data-bot and finance-agent. Each agent can only invoke tools within its registered capabilities. The test scenarios below trace requests through these delegation paths.

The following policies are deployed to the same Verified Permissions policy store. In the reference implementation, policy files are located under cedar/policies/ organized by layer: layer1-agent-to-tool/, layer2-agent-to-agent/, and layer3-originating-user-auth/.

Layer 1 (agent-to-tool): This policy permits the finance-agent to invoke the process_payment tool only when three conditions are met: the agent’s trust score is at least 3, it belongs to the payments namespace, and it’s deployed in the production lifecycle stage. If any condition fails, the request is denied. The agent’s trust_level, namespace, and lifecycle_stage aren’t self-reported in a production deployment. Instead, the evaluator retrieves these attributes from the Verified Permissions entity store using the agent_id as a lookup key.

Important: The reference implementation accepts these values from the request payload for simplicity. Production deployments must validate agent attributes against an authoritative source to prevent a compromised agent from escalating its own trust.

The trust_level attribute uses a 1–5 integer scale that represents an agent’s verified maturity: 1 for newly registered and untested agents, 3 for agents that have passed integration testing and security review, and 5 for agents with a proven production track record. Organizations assign trust levels through their agent promotion pipeline, not through self-declaration. The lifecycle_stage attribute (development, staging, production) prevents pre-production agents from invoking production tools, even if they have the correct namespace and trust score.

// L1-001: Finance agent can invoke payment tools
permit(
  principal == AgentAuthz::Agent::"finance-agent",
  action == AgentAuthz::Action::"invoke_tool",
  resource == AgentAuthz::Tool::"process_payment"
) when {
  principal.trust_level >= 3 &&
  principal.namespace == "payments" &&
  principal.lifecycle_stage == "production"
};

Layer 2 (agent-to-agent delegation) enforces depth limits and capability constraints. The orchestrator agent delegates tasks to data-bot only when the delegation chain is three hops or fewer and the requested capabilities are a subset of data-bot’s registered capabilities. A separate forbid policy (L2-004) enforces a hard system-wide limit of five hops regardless of which agents are involved.

// L2-002: Orchestrator can delegate to data agent
permit(
  principal == AgentAuthz::Agent::"orchestrator",
  action == AgentAuthz::Action::"delegate_task",
  resource == AgentAuthz::Agent::"data-bot"
) when {
  context.delegation_depth <= 3 &&
  context.target_capabilities.containsAll(context.requested_capabilities)
};

Layer 3 (originating user authorization) keeps the agent as the principal, but the policy evaluates context.originating_user to validate the human who initiated the request. data-bot invokes the delete_records tool only when the originating user has the admin role, has verified MFA, and the delegation chain is at most two hops deep. Without this layer, an agent with the right capabilities could invoke destructive tools regardless of who initiated the request.

// L3-001: High-risk tool (delete_records) requires admin + MFA
permit(
  principal == AgentAuthz::Agent::"data-bot",
  action == AgentAuthz::Action::"invoke_tool",
  resource == AgentAuthz::Tool::"delete_records"
) when {
  context.originating_user.role == "admin" &&
  context.originating_user.mfa_verified == true &&
  context.delegation_depth <= 2
};

Key design point: The principal remains the agent, not a user entity. The user’s role and MFA status are checked through context attributes, keeping the schema to two entity types and two actions.

The reference implementation uses Amazon Cognito with TOTP MFA, but most OIDC-compliant providers (Okta, Microsoft Entra ID, Auth0, or AWS IAM Identity Center) work with this pattern. The authentication-to-signing flow is described in the preceding Authentication before authorization section. To use a different IdP, replace the Cognito authorizer on API Gateway with a Lambda or JWT authorizer for your IdP’s issuer URL. Cedar policies remain unchanged.

The reference implementation deploys five CloudFormation stacks: KmsStack, VerifiedPermissionsStack, LambdaStack, SecurityLakeStack, and MonitoringStack. The following commands deploy the stacks in dependency order:

cdk deploy KmsStack -c account_id=YOUR_ACCOUNT_ID -c guardrail_id=YOUR_GUARDRAIL_ID
cdk deploy VerifiedPermissionsStack -c account_id=YOUR_ACCOUNT_ID -c guardrail_id=YOUR_GUARDRAIL_ID
cdk deploy LambdaStack -c account_id=YOUR_ACCOUNT_ID -c guardrail_id=YOUR_GUARDRAIL_ID
cdk deploy SecurityLakeStack -c account_id=YOUR_ACCOUNT_ID -c guardrail_id=YOUR_GUARDRAIL_ID
cdk deploy MonitoringStack -c account_id=YOUR_ACCOUNT_ID -c guardrail_id=YOUR_GUARDRAIL_ID

Three end-to-end scenarios validate the evaluation model across different user roles, MFA states, and delegation depths. To run the tests:

1. Set the API endpoint from the deployment output:

export API_ENDPOINT=$(aws cloudformation describe-stacks --stack-name LambdaStack \
  --query "Stacks[0].Outputs[?OutputKey=='ApiEndpoint'].OutputValue" --output text)

2. Run the end-to-end tests:

.venv/bin/python -m pytest tests/e2e/ -v -s

The end-to-end tests cover the three scenarios described in the following sections. Each test sends a request through the deployed API and validates the per-layer authorization decisions.

A support-role user (no MFA) requests record deletion through orchestrator and data-bot.

Without Layer 3, this request would have been permitted based on agent capabilities alone, demonstrating why originating user authorization is essential.

An admin user with MFA requests the same operation through the same chain.

An admin with MFA requests the same operation, but the delegation chain has six hops. This scenario tests the Layer 2 depth constraint independently of user authorization.

Even an authorized admin can’t bypass the delegation depth constraint.

The AWS Office of the CISO published Four security principles for agentic AI systems. The following table shows how this solution maps to each principle.

Each evaluation produces an OCSF 99001 audit event with request ID, user identity, delegation chain, per-layer decisions, and latency.

The following table maps this implementation to NIST SP 800-53 Rev. 5 controls. Customers are responsible for evaluating whether it meets their compliance requirements.

Deploy the Cedar policy store in a central security account and use cross-account IAM roles for workload accounts to call verifiedpermissions:IsAuthorized. Use AWS Organizations service control policies (SCPs) to prevent workload accounts from creating their own policy stores. For standardizing user identity attributes across the organization, consider IAM Identity Center or a centralized OIDC provider that issues consistent claims to your workload accounts. This helps ensure that the context.originating_user attributes are uniform across accounts and agents.

For production deployments, consider extending this pattern with human-in-the-loop escalation for borderline denials, multi-tenant Cedar policy isolation, and Amazon Simple Storage Service (Amazon S3)-backed dynamic policy hot-reload for emergency tool shutdowns.

To avoid ongoing charges, delete the deployed resources:

cdk destroy MonitoringStack SecurityLakeStack
cdk destroy LambdaStack
cdk destroy VerifiedPermissionsStack
cdk destroy KmsStack
aws logs delete-log-group --log-group-name /cedar-evaluator/audit  # if RETAIN policy

Multi-agent AI systems need authorization boundaries at every delegation hop. The three-layer Cedar policy model with OAuth 2.0 authentication provides that protection while maintaining least-privilege access. Combining a trusted IdP (AuthN) with Cedar policy evaluation (AuthZ) creates an authorization boundary around each tool invocation, verifying agent capability (L1), delegation path (L2), and originating user authority (L3). The pattern works with an OIDC-compliant IdP and a compute platform that can call Amazon Verified Permissions. Clone the reference implementation and adapt the Cedar policies to your organization’s requirements. For more information, see the Cedar policy language documentation and the Amazon Verified Permissions User Guide.

• Cedar policy language documentation
• Amazon Verified Permissions User Guide
• OWASP Top 10 for Agentic Applications
• Four security principles for agentic AI systems
• OAuth 2.0 Token Exchange (RFC 8693)
• NIST AI Agent Standards Initiative
• NIST SP 800-53 Rev. 5
• OCSF schema documentation

If you have feedback about this post, submit comments in the Comments section below.