Briefing · March 2026
Keeping staff work inside the organisation’s approved perimeter.
The first failure mode we see in unsupervised tool use is leakage of material that the organisation never intended to leave its tenancy. Training that ignores that failure mode teaches fluency and increases exposure.
Enterprise enablement starts with a written list of tools the organisation has licensed and configured. That list is read aloud in the opening hour. Participants are shown where the enterprise instance sits, how identity is bound to the corporate directory, and which logging the vendor already provides. We do not train on consumer accounts, even when a participant already uses them at home.
The second list is a data classification. Typical classes include public, internal, confidential and restricted. For each class we state whether it may enter a prompt, whether it may be used as retrieval context, and whether an output that restates it may be stored. The exercise is dull by design. Staff who can recite the classes make fewer improvisations later.
A third list covers retention. Some organisations forbid the saving of prompts. Others require a ticket number on every substantial generation that will be used in a customer file. We do not invent a single rule. We teach the rule the client already owns, and we assess whether participants can apply it to three sample tasks from their own function.
Where a client has not yet written those lists, discovery pauses the design of exercises. A programme that invents policy in the room creates conflict with legal and security teams after the fact. The briefing for sponsors is therefore short: supply the lists, or accept a governance-literacy module before any applied practice. See Governance for the module outline.
Facilitators use the client’s screens. Participants complete tasks on the same tenancy they will use on Monday morning. Screenshots in workbooks are taken from that tenancy after permission is recorded. This removes the common gap between a polished demo environment and the locked-down instance staff actually receive.
The assessed outcome for this theme is a short written judgement: given a document class and a task, may the participant use the tool, and what must be redacted first. Pass marks are agreed with the sponsor before the cohort opens. Failures return to a supervised clinic rather than to an unsupervised desk.
