Amazon Bedrock Agents — Agent Audit

ESTIMATED SCORE — NOT VERIFIED

This assessment is based solely on publicly available documentation, marketing materials, and product announcements. Amazon Bedrock Agents has not submitted to AgentGoverning for independent adversarial verification. This score is an estimate only and may not reflect actual platform capabilities. Amazon Bedrock Agents is invited to submit for formal assessment at framework@agentgoverning.com. Estimated scores carry no certification status.

How This Score Is Calculated

This estimated score reflects publicly documented agent deployment governance capabilities assessed across 10 capability categories against the 508 Agent Audit dimensions of AGS v2.2:
• Mandate & autonomy (27 dims)
• Agent orchestration (65 dims)
• Trust & identity (30 dims)
• Detection & containment (35 dims)
• Financial controls (45 dims)
• Human oversight (7 dims)
• Memory & knowledge (18 dims)
• Sector-specific (180 dims)
• Other core governance (93 dims)
• Deployment & lifecycle (8 dims)

Each category scored 0–100% based on public documentation coverage. Overall score = weighted average across all 10 categories.

Based solely on publicly available documentation as of April 2026.

What This Score Means

Amazon Bedrock Agents scores highest among competitors on the Agent Audit due to its Cedar declarative policy language and AgentCore Gateway — a genuine deterministic enforcement layer operating outside the LLM reasoning loop. The supervisor/sub-agent architecture provides real multi-agent orchestration. However, agent-specific governance capabilities beyond Cedar policy enforcement are limited: no inter-agent trust attestation, no graduated autonomy framework, no delegation depth governance, and no cryptographic state sealing. The majority of Agent Audit dimensions require structural enforcement mechanisms beyond policy-level controls..

Mandate

55%

Orchestration

30%

Trust

20%

Detection

20%

Financial

15%

Oversight

15%

Memory

15%

Sector

20%

Core

25%

Deployment

20%

Key Strengths

AG-001

Operational Boundary Enforcement

Cedar declarative policy language provides genuine deterministic enforcement of operational boundaries outside the LLM reasoning loop via AgentCore Gateway.

Score: 2 / 3

AG-057

Delegation Chain Governance

Supervisor/sub-agent architecture provides real multi-agent orchestration with structured delegation patterns.

Score: 1 / 3

AG-060

Deployment Gates

AWS deployment infrastructure provides basic deployment gate controls through standard AWS service deployment mechanisms.

Score: 1 / 3

The following gap analysis is based on publicly available documentation only. These are estimated structural gaps, not verified findings. Amazon Bedrock Agents may have implemented controls not visible in public documentation.

Critical Gaps

AG-058

Inter-Agent Trust Attestation

No inter-agent trust handshake or attestation framework. Agents cannot cryptographically verify the governance status of other agents before delegation.

Score: 0 / 3 — Structurally Absent

AG-786

Cryptographic State Sealing

No cryptographic seal or tamper-resistance mechanism for agent state. Agent state integrity cannot be independently verified.

Score: 0 / 3 — Structurally Absent

AG-788

Federated Broadcast Governance

No federated broadcast mechanism for governance policy propagation across distributed agent deployments.

Score: 0 / 3 — Structurally Absent

Full Dimension Assessment

Area	Strongest Capabilities	Status
Strongest Documented Capabilities
Mandate	Mandate boundary enforcement (Cedar + AgentCore Gateway)	Evidenced
Orchestration	Multi-agent orchestration (supervisor/sub-agent architecture)	Evidenced
Identity	Agent identity (IAM)	Evidenced
Documentation	Provider documentation	Evidenced
Structurally Absent Capabilities
Trust	Inter-agent trust handshakes	Absent
Trust	Governance passports	Absent
Oversight	Graduated autonomy	Absent
Orchestration	Delegation chain governance	Absent
Core	Cryptographic seal resistance	Absent
Core	Federated broadcast	Absent
Core	Composite scoring	Absent
Memory	Memory/RAG governance	Absent
Financial	Agent-level financial crime	Absent

Legal Disclaimer. This assessment is an independent estimate based on publicly available documentation and does not constitute legal, compliance, or certification advice. AgentGoverning is not affiliated with, endorsed by, or sponsored by Amazon Web Services, Inc. or its affiliates. "Amazon Bedrock Agents", "Cedar", "AgentCore", and "AWS" are trademarks of Amazon.com, Inc. or its affiliates. All scores are estimated and unverified. This page does not constitute a certification, endorsement, or guarantee of any kind. The assessment methodology, scoring criteria, and dimension definitions are the intellectual property of AgentGoverning and are licensed under CC BY 4.0. For formal verification, platforms must submit to independent adversarial assessment. Estimated scores may change as new documentation becomes available. AgentGoverning accepts no liability for decisions made based on estimated scores.