Responsible AI Center

Plain-English guardrails for using AI safely, reviewing outputs, minimizing data exposure, and avoiding common failure modes.

Use approved data only. Treat AI output as a draft and verify facts.

Data classification

Use AI with approved data, verify important claims, and document human review for high-impact workflows.

Secrets management

Use AI with approved data, verify important claims, and document human review for high-impact workflows.

Prompt injection

Use AI with approved data, verify important claims, and document human review for high-impact workflows.

Hallucinations

Use AI with approved data, verify important claims, and document human review for high-impact workflows.

Human review

Use AI with approved data, verify important claims, and document human review for high-impact workflows.

Compliance-aware AI usage

Use AI with approved data, verify important claims, and document human review for high-impact workflows.

AI-generated code risks

Use AI with approved data, verify important claims, and document human review for high-impact workflows.

Data minimization

Use AI with approved data, verify important claims, and document human review for high-impact workflows.

Model selection

Use AI with approved data, verify important claims, and document human review for high-impact workflows.

Auditability

Use AI with approved data, verify important claims, and document human review for high-impact workflows.

Should I use AI for this?

Step 1: Is the data approved for this tool?
Step 2: Can a human review the output?
Step 3: Would a wrong answer create customer, legal, safety, or employee impact?
Step 4: Can you test with synthetic or minimized data first?

Safe / Caution / Do Not Use Matrix

Safe

Brainstorming public campaign ideas, drafting internal outlines, summarizing approved non-sensitive content.

Caution

Using internal data, drafting customer-facing communications, analyzing decisions that affect people.

Do Not Use

Pasting secrets, credentials, regulated records, private employee data, or unapproved customer-sensitive data.

Vibe Coding Safety Guide

Never paste secrets or credentials.
Review generated code before using it.
Add tests and run them.
Understand dependencies and validate licenses.
Watch for insecure patterns.
Do not blindly deploy generated code.
Use least-privilege tokens.
Document human ownership.