Scope and purpose
Before building or purchasing an AI system, clarify its intended use, affected groups and who is accountable for its outputs. A system should not quietly expand into higher-risk uses without fresh review.
Data and security
Collect what is necessary, limit access, secure credentials and understand where data is processed. Projects involving sensitive information require appropriate agreements and technical safeguards.
Evaluation and oversight
Assess performance on realistic examples, investigate error patterns, document limitations and provide a way to challenge or override consequential outputs. Monitor whether performance changes over time.
Our boundary
These principles do not certify any service as compliant with a particular law or standard. Controls must be assessed against the actual system, jurisdiction and use case. See the site privacy notice for further information.