What You Need to Know
The sanctioned workspace is the Claude workspace your organisation provides. A personal account is one an individual holds for themselves. Shadow AI is work material in an AI tool the organisation does not provide. The rule is short: use the workspace your organisation provides. It sounds like etiquette. It has substance, and seeing it lets you answer any version without memorising a table.
A minicab firm provides Claude to its office staff. A new dispatcher's access has not come through, and a hotel contract is due Friday. She pastes the contract and the complaints log into her personal Claude account because it is quicker. The bookings manager finds out.
Why the Sanctioned Workspace Is Substance
When an organisation provides Claude, the arrangement governs how its data is handled. The organisation decides what happens to its data. Anthropic handles it only on the organisation's instructions. In the Privacy Center's terms, the customer is Controller of the data its users submit, and Anthropic is Processor on the customer's behalf. The commercial training article states that by default Anthropic will not use inputs or outputs from commercial products to train its models. Administrators hold the workspace's settings. On the Enterprise plan, the organisation sets custom retention timelines for its data. Users still delete their own chats from their visible history. The organisation holds the workspace's retention, not the deletion action. Agreement, administration and data handling all attach to the sanctioned workspace, and that is what the rule protects. In the firm's workspace, the contract would sit under the firm's agreement.
Personal Accounts Are Different
A personal account sits under the individual's own consumer relationship with Anthropic. Their own settings decide whether chats feed model improvement. Consumer deletion and retention mechanics apply. The employer does not administer, see or govern any of it. The problem is not carelessness. The organisation's agreement and controls do not reach the account. The dispatcher's contract now sits under her agreement, not the firm's. A paid personal plan with careful settings would change nothing, because the question is whose agreement governs the data.
The Shadow-AI Response Has Two Halves
The response to a colleague's workaround has two halves. First, do not copy the workaround, however pressing the deadline. Second, raise it with whoever owns the policy or the data. The disclosure has already happened, and someone with authority to deal with it needs to know. An unreported workaround also continues. The colleague's underlying problem is real: access has not arrived for work that is due. Only the organisation can fix it. Raising it is how the organisation finds out its access process is too slow, so it is not informing on a colleague. The bookings manager keeps her own work in the workspace and tells the policy owner about the disclosure and the late access. Asking the dispatcher to delete her chats or close her account is after-the-fact tidying, because the contract has already left the firm's hands.
Policy Silence and Working Habits
A policy's silence means the question has not been answered. The professional move is to ask the policy's owner. This is the escalation objective 6.2 teaches, an action rather than a refusal. If the firm's policy says nothing about pasting complaints into Claude, the bookings manager asks its owner first. The habits underneath are unglamorous and examinable:
- know which account and plan you work in
- know who administers it
- follow the rules for what may be uploaded
- know how to report a problem
When policy conflicts with convenience, the policy wins. When the policy is genuinely wrong or too slow, get its owner to change it rather than work around it.
In the exam
The recurring stem is a colleague using a personal account for work material, because it is quicker or their access has not come through. Candidates who treat the rule as etiquette lose marks to distractors built on good intentions. The keyed answer does both halves, and candidates reliably miss the second. Distractors offer the first half alone, or deleting chats and closing the account as remediation, or a paid personal plan with careful settings. The most common distractor treats anything the policy does not mention as allowed.
Boundary discipline
Whether material may be uploaded at all is objective 6.2. Designing a process is Domain 4. Escalation here goes to whoever owns the policy or the data, never to a technical specialist.
Anthropic publishes no guidance on what an organisational AI policy should contain. This objective rests on the exam guide's wording and on the documented product differences behind the sanctioned workspace.
Exam Traps
Fixing only half of the shadow-AI situation
Declining to copy the workaround while saying nothing leaves the disclosure unknown and the access gap unfixed, so the workaround continues. The keyed answer does both halves: stop the practice and raise it with whoever owns the policy or the data.
Offering deletion as remediation for material already disclosed
This is the technical placebo in governance dress. Having the colleague delete conversations or close the account is after-the-fact housekeeping; the material already left the organisation's hands, and nobody with authority has been told.
Treating policy silence as permission
A policy that does not mention something has not answered the question. Assuming permission converts a gap in a document into an unauthorised disclosure; the keyed move is asking the policy owner, which is an action, not a refusal.
Practice Scenario
The operations lead at a recycling social enterprise discovers that a colleague has been pasting the local authority contract and staff shift records into a personal free Claude account at home, because the organisation's licences have not been issued yet and the work is due. The enterprise's data rules keep its material out of personal accounts on any plan, and any slip must be reported. What should the operations lead do?
Build Exercise
Map Your Own Governance Position
What you'll learn
- How to establish which account and plan actually govern your work material
- How to locate your AI policy and its owner
- How to surface the gaps between the policy and your normal week
- How to prepare the shadow-AI conversation before you need it
- Write down which Claude account you actually use for work material: personal or organisation-provided, which plan, and who, if anyone, administers it
Why: Everything in this objective attaches to the account, so knowing which one you are in is the foundation habit.
You should see: Three facts on one line: account type, plan, administrator.
- Find the AI policy of your organisation; if you cannot find one, or cannot name who owns it, that is the finding
Why: A policy you cannot locate cannot guide your decisions, and discovering the gap is itself the governance result.
You should see: Either the policy document and the name of its owner, or a written note that neither could be found.
- List what the policy says may and may not be uploaded, and mark anything you handle in a normal week that the policy does not mention
Why: The silent areas are where policy-silence-as-permission would bite you, so they need surfacing before the pressure moment.
You should see: Two lists: what the policy answers, and what it is silent on.
- For one silent area, draft the two-sentence question you would send the policy owner, and send it
Why: Asking the owner is the escalation this domain teaches, and doing it once in calm conditions makes it available under pressure.
You should see: A sent message naming the material, the intended use, and asking whether the policy permits it.
- Write down how you would raise a workaround by a colleague: what you would say to the colleague first, and who you would tell
Why: The two-half response fails in practice when the second half has no named recipient, so decide both in advance.
You should see: Two lines: the sentence to the colleague, and the named person who owns the policy or the data.
Sources
- Does Anthropic Act as a Data Processor or Controller? — Anthropic
- Is my data used for model training? (commercial products) — Anthropic
- Is my data used for model training? (consumer products) — Anthropic
- How do I change my model improvement privacy settings? — Anthropic
- Can you delete data that I sent via Team and Enterprise plans? — Anthropic
- Claude Certified Associate – Foundations Exam Guide — Anthropic. Anthropic publishes no guidance on what an organisational AI policy should contain. This lesson draws on the exam guide's objective and the product differences documented above.