Policy position
Usual Industries requires the responsible use of approved AI services for designated work and prohibits their use outside approved boundaries. A tool being installed, available in a browser or recommended by a colleague does not constitute approval. The person using AI remains the author, reviewer and accountable owner of what the company sends or does.
The three decisions before a prompt
First identify the information being entered. Then identify the tool, account and permitted use. Finally identify the human who will verify the output before anyone relies on it. If one of these cannot be named, stop at the first decision. A request to “make this sound more official” does not relax any of the three.
Where AI is required
Where an approved workflow includes an AI first pass, colleagues must use or document why the task is exempt. Quietly ignoring an approved workflow can be investigated just as surely as putting restricted material into an unapproved tool. Adoption is a managed obligation, not a competition to generate the most text.
Where AI is prohibited
Do not submit confidential, personal, privileged, transaction-sensitive or government-restricted material to a tool that has not been expressly cleared for that information. Do not allow AI to make a final employment, financial, legal, safety or customer decision. Generated assertions, figures, citations and quotations must be checked against the underlying record.
Joint oversight
The AI Center of Excellence maintains the tool register and use-case inventory. IT Security controls access and investigates technical exposure. Legal assesses privilege, commitments and disclosure. Internal Audit tests evidence, exceptions and the accuracy of attestations. A use case may require all three reviews even when it began as a request for a better slide title.
Consequences
Deliberate misuse, concealment or repeated failure to review outputs can lead to loss of tool access, corrective training, removal from a role, disciplinary action up to termination and referral where required. Repeated failure to perform a required approved AI step can also trigger review. Severity depends on conduct, impact and whether the colleague promptly reported a problem; a fabricated citation that reaches the Board receives more attention than an abandoned draft.
