Ask KJ
How can companies adopt AI securely?
Classify the information the system can process, limit access and actions, test likely failures, and assign ownership for review and incident response.


Browse related state and city resources
How to approach it
Start with rules for which data can go into which tools. Evaluate vendor retention, permissions, and access logs. Tools that read external content should treat it as untrusted, especially when an AI assistant can invoke tools or publish information.
Practical example
An internal AI assistant can draft a response using approved knowledge sources. Keep its data access scoped to the user. Require review before sending messages or making changes to business records.
Costs, limitations, and next steps
Security effort rises with integrations and sensitive information. Use a limited pilot, test revocation and recovery, and review obligations with the appropriate organization specialists.
Reference
NIST Generative Artificial Intelligence Profile
