Classify risk before use
Consider the possible harm if the output is wrong, biased, disclosed or acted upon without review. Apply stronger controls to employment, lending, health, legal, safety and financial decisions.
- Identify affected people
- Document prohibited uses
- Require qualified human review
Protect data and rights
Use only information the business is authorized to process. Review tool terms, retention, training use, access and security before uploading sensitive material.
- Minimize personal data
- Respect copyright and permissions
- Keep confidential material out of unapproved systems
Monitor and correct
Record important AI assisted processes, test results and incidents. Give employees and customers a way to report errors and obtain human review.
- Audit high impact outputs
- Update rules as tools change
- Stop a workflow when controls fail
