Only after the organization has approved that use and confirmed the relevant data, contractual, security, and privacy requirements. Check the provider's applicable terms and settings, including retention, training use, access, and subprocessors. Do not assume a paid account or an enterprise label answers every question. Prefer a limited or de-identified input when it meets the business need, and verify that removing obvious names has not left identifying details. Give employees an approved route for the task. If sensitive information was already entered into an unapproved service, use the incident process rather than quietly deleting the chat and assuming the issue is resolved.
Related: SOC 2 audit and compliance hub.
Further reading: NIST Generative AI Profile.