AI Change Impact Assessment

Use this template for high-risk AI/system changes, including changes to:

• Agent permissions, approval modes, shell/file/network tools, or policy enforcement.
• Prompt behavior, context selection, model/provider defaults, or provider request handling.
• Audit logs, report generation, data minimization, or release security.

Copy the template into the pull request description or into a dated file under docs/governance/assessments/ when the assessment needs to be retained as a standalone record.

Change Summary

• PR / issue:
• Owner:
• Date:
• Change class: A2 / A3
• Components changed:

Intended Use

• What user workflow does this change support?
• Which modes are affected: Chat / Cowork / Code / terminal / web / server?
• Does the change alter default behavior?

Stakeholders

• Maintainer:
• User or organization admin:
• Security reviewer:
• Affected third-party providers:

Data and Privacy

• What user data can be read, transformed, stored, or sent to a provider?
• Are prompts, provider responses, file contents, API keys, or secrets stored?
• Does .dvalincodeignore or policy enforcement still apply?
• Does audit logging remain minimized?

Autonomy and Permission Impact

• Does the change expand file write, delete, shell, network, model, or tool access?
• Does it change when user approval is required?
• Does it change Plan/Ask/Auto/Bypass behavior?
• What prevents prompt injection from escalating permissions?

Model and Supplier Impact

• Are new model providers, endpoints, SDKs, or hosted services introduced?
• Are provider requests routed through the governed egress path?
• Are provider errors, rate limits, and retries safe and observable?

Security and Supply Chain

• Are dependencies added or updated?
• Are GitHub Actions pinned to commit SHAs?
• Do CI, CodeQL, Scorecard, and release verification still run?
• Are new secrets required? If yes, where are they stored and scoped?

Testing and Evidence

• Unit tests:
• Integration or e2e tests:
• Manual verification:
• Audit/report verification:
• Screenshots or logs:

Residual Risk

• Remaining risks:
• Accepted exceptions:
• Owner:
• Review date:

Decision

• [ ] Approved
• [ ] Approved with conditions
• [ ] Rejected

Reviewer: Date: