Usually yes, for the records your own people produced. What decides it is not whether AI training is lawful. It is whether you own the specific material, and a lot of what sits in a company's enterprise data is somebody else's.
Owning the file is not owning the rights
17 U.S.C. 202 separates the two. Possessing a document, or buying the hard drive it sits on, conveys no copyright in it. An estate that buys a company's servers has the files and not the right to license what a third party wrote. This is the single most common reason a corpus cannot be sold.
What you almost certainly can license
Work your employees produced in the course of their employment. Under the work made for hire doctrine that is yours from the moment it is written: internal mail, drawings, quotes, reports, specifications, the contents of your own systems.
What needs checking first
Contractor and agency work, where the rights follow the contract and often stayed with them. Licensed software, fonts, stock imagery and third party data feeds. Material your customer owns, which on most professional services drives is a large share of the folder. Anything carrying a confidentiality obligation you signed.
Refused whatever the price
Substance use and behavioral health treatment records including employee assistance files, under 42 CFR Part 2, which has no de-identification exception. HIV records. Medical and clinical records about any person, workers compensation and health claims included. Privileged client files. Classified and export-controlled material. Anything under a litigation hold. These are refusals, not discounts.
People in the records
Staff, customers and suppliers are named all through real enterprise data. They are licensed as held. You confirm you may share what you upload. Each buyer handles the personal data under its own legal duties. Health records and substance use treatment records are excluded.