AI Agent Impact Assessment Template¶

Purpose: Structured template for assessing risks and impacts before deploying AI agents in high-risk environments. Aligns with Colorado AI Act (SB 21-169), EU AI Act Article 9, and NIST AI RMF MAP function.

Usage: Complete this template for each agent deployment. Store completed assessments alongside the agent's policy YAML in version control.

1. Agent Identification¶

Field	Value
Agent Name
Agent DID	`did:agentmesh:...`
Version
Owner / Sponsor
Deployment Environment	☐ Development ☐ Staging ☐ Production
Assessment Date
Assessor
Review Due Date

2. Purpose and Scope¶

2.1 Agent Purpose¶

Describe what the agent does, what decisions it makes, and what actions it takes.

2.2 Target Population¶

Who is affected by the agent's actions? Include both direct users and downstream affected parties.

2.3 Decision Types¶

Decision	Consequential?	Reversible?
e.g., Loan approval	☐ Yes ☐ No	☐ Yes ☐ No
e.g., Content moderation	☐ Yes ☐ No	☐ Yes ☐ No

Consequential decisions (Colorado AI Act): decisions that have a material legal or similarly significant effect on consumers in education, employment, financial services, healthcare, housing, insurance, or government services.

3. Risk Classification¶

3.1 Risk Level¶

Framework	Classification	Justification
EU AI Act	☐ Minimal ☐ Limited ☐ High ☐ Unacceptable
NIST AI RMF	☐ Low ☐ Moderate ☐ High ☐ Critical
OWASP Agentic	Applicable risks:
Organization	☐ Tier 1 ☐ Tier 2 ☐ Tier 3

3.2 OWASP Agentic Top 10 Assessment¶

#	Risk	Applicable?	AGT Mitigation
ASI-01	Agent Hijacking	☐ Yes ☐ No	Policy engine, prompt injection detection
ASI-02	Tool Misuse & Exploitation	☐ Yes ☐ No	Capability allow/deny lists
ASI-03	Tool Poisoning	☐ Yes ☐ No	MCP security scanner
ASI-04	Insecure Data Handling	☐ Yes ☐ No	Attribute ratchets, DLP
ASI-05	Insecure Output	☐ Yes ☐ No	Pre-output policy stage
ASI-06	Confused Deputy	☐ Yes ☐ No	Zero-trust identity, trust scoring
ASI-07	Insecure Inter-Agent Communication	☐ Yes ☐ No	E2E encryption (Signal protocol)
ASI-08	Cascading Agent Failures	☐ Yes ☐ No	Circuit breaker, kill switch
ASI-09	Inadequate Logging	☐ Yes ☐ No	Tamper-evident audit, OTel
ASI-10	Resource Exhaustion	☐ Yes ☐ No	Rate limiting, token budgets

4. Data Assessment¶

4.1 Data Inputs¶

Data Source	Contains PII?	Classification	Jurisdiction
	☐ Yes ☐ No	☐ Public ☐ Internal ☐ Confidential ☐ Restricted

4.2 Data Outputs¶

Output Type	Contains PII?	External Recipients?
	☐ Yes ☐ No	☐ Yes ☐ No

4.3 Data Residency¶

Storage locations:
Cross-border transfers: ☐ Yes ☐ No
If yes, legal basis:

5. Bias and Fairness¶

5.1 Protected Classes Potentially Affected¶

☐ Race/Ethnicity ☐ Gender ☐ Age ☐ Disability ☐ Religion ☐ National Origin ☐ Other: ___

5.2 Fairness Testing¶

Has the agent been tested for disparate impact? ☐ Yes ☐ No
Testing methodology:
Results summary:

5.3 Mitigation Measures¶

Describe controls to prevent discriminatory outcomes.

6. Governance Controls¶

6.1 AGT Policy Configuration¶

# Reference the agent's policy file
policy_file: policies/<agent-name>.yaml
extends:
  - policies/org-baseline.yaml

6.2 Approval Gates¶

Action	Requires Approval?	Approver
e.g., Financial transaction > $10K	☐ Yes ☐ No

6.3 Human Oversight¶

Kill switch configured? ☐ Yes ☐ No
Human-in-the-loop for consequential decisions? ☐ Yes ☐ No
Escalation path defined? ☐ Yes ☐ No

7. Transparency and Explainability¶

Can the agent explain its decisions to affected individuals? ☐ Yes ☐ No
Is there a consumer-facing disclosure that AI is being used? ☐ Yes ☐ No
Are audit logs accessible for review? ☐ Yes ☐ No

8. Incident Response¶

Incident classification defined? ☐ Yes ☐ No (see incident-response-workflow.md)
Rollback procedure documented? ☐ Yes ☐ No
Notification process for affected individuals? ☐ Yes ☐ No

9. Sign-Off¶

Role	Name	Date	Signature
Agent Owner
Security Review
Privacy Review
Legal/Compliance
Management Approval

10. Review Schedule¶

Next review date:
Review trigger events: ☐ Major version change ☐ Policy update ☐ Incident ☐ Regulatory change ☐ Annual

Related: NIST AI RMF Alignment · EU AI Act Checklist · SOC 2 Mapping · Record Retention Policy