Shadow AI at the first gate
Hartwell Trust Bank checks an outside AI use case with fictional data.
Coming
See every demonstrationFREE CHECK
Private AI readiness
Choose the data and operating boundary before selecting a model or device.
01 Source02 Boundary03 Human decision04 Evidence
WHAT TO REVIEW
Four questions for the team that owns the decision.
- Which data categories must stay in a defined boundary?
- Can the workload run on representative sample data?
- Who owns patching, evaluation, and incident response?
- Have you compared tenant-private and local operation?
Take the PDF with you.
Enter a work email, company and role to request the PDF. This action records an SIT inquiry for follow-up. Marketing emails require a separate choice.
Download the PDF directlyTHE CHECK
What does the private AI readiness check test?
Which data categories must stay in a defined boundary, whether the workload can run on representative sample data, who owns patching, evaluation and incident response, and whether tenant-private and local operation have been compared.
Four questions on the data and operating boundary before you pick a model or a device, with a PDF you can download directly and no email required.
THE MICROSOFT INSIGHT
Is the data boundary the same for every Microsoft AI feature?
No: models provided by Anthropic as a Microsoft subprocessor, used in Copilot experiences including Cowork, are currently excluded from the EU Data Boundary, so the boundary decision is made feature by feature.
Models provided by Anthropic as a Microsoft subprocessor, used in Copilot experiences including Cowork, are currently excluded from the EU Data Boundary, and tenants in the EU and the UK have them disabled by default. (Microsoft Learn, checked Oct 31, 2026)
Decide feature by feature and the general counsel gets a table. Decide once for everything and the general counsel gets a surprise.
THE METHOD
How does the check become a boundary decision?
Each answer fixes one side of the boundary: the data that must stay in, the sample data to test with, the people who run it, and the two operating options compared.
- Data: the categories that must stay inside a defined boundary, named.
- Sample: the workload can run on representative sample data before any real record.
- Owners: patching, evaluation and incident response each have a name.
- Comparison: tenant-private and local operation compared on the same workload.
THE DEVICE
Does private AI mean the firm has to buy hardware?
Not by default: the tenant-private path inside Microsoft 365 and Azure is tested first, and a local or dedicated option is chosen only where the data rules require it.
Microsoft Defender for Cloud Apps lists generative AI apps as a category, and an app tagged Unsanctioned is blocked on devices onboarded to Microsoft Defender for Endpoint, with up to three hours of propagation. (Microsoft Learn, checked Oct 31, 2026)
Test the boundary first and the hardware decision is a line in a plan. Buy the hardware first and the boundary decision is made by the box.
THE NEXT RUNG
What comes after the check?
A one-hour private AI briefing in plain words, then a one-day workshop that fixes the boundary, then a four-week pilot inside it.
Each is its own scope, quoted after a scoping call, and nothing is widened until the firm decides in writing.
AFTER TWO WEEKS
What will be working after two weeks?
- You hold the PDF and the four sides of the boundary are marked tested or assumed.
- The data categories that must stay inside are named.
- The owners of patching, evaluation and incident response are named, or their absence is the finding.
- You have booked, or decided not to book, the briefing.
HONEST ABOUT LIMITS
What this does not do yet
It does not cover personal devices the firm does not manage. Microsoft Defender for Cloud Apps lists generative AI apps as a category, and an app tagged Unsanctioned is blocked on devices onboarded to Microsoft Defender for Endpoint, with up to three hours of propagation. (Microsoft Learn, checked Oct 31, 2026) Blocking works on devices onboarded to Microsoft Defender for Endpoint, so the boundary includes a device decision.
It does not pick a model or a device: that waits for the workshop and the comparison on your workload.
It does not create a lead unless you ask for the PDF by email and say so.
THE ONE THING TO REMEMBER
Choose the AI boundary before choosing hardware, and choose it on the data rules.
Said another way: the question is where the data may go, not which model is cleverest.
Said a third way: feature by feature, with the Learn source and the date.
SHOW YOUR WORK
See the control path on a fictional bank.
DIRECT ANSWERS
Questions before you decide.
Who should take this check?
The executive accountable for client confidentiality and the person who runs IT, together. Answers that differ between them are the finding.
Is the PDF gated?
No. Download it directly. Asking for it by email is optional and creates an inquiry; marketing email is a separate box.
What is tenant-private operation?
AI that runs inside your Microsoft 365 and Azure tenant under your existing permissions and labels, as opposed to a model and device you operate yourself.
What comes after the check?
A one-hour briefing in plain words, a one-day workshop that fixes the boundary, and a four-week pilot inside it, each its own scope.
How is it priced?
Quoted after a scoping call. The briefing, the workshop and the pilot are separate scopes.
NEXT STEP
Start with a boundary you can explain.
Choose a focused check, or speak with the team about the workflow you need to improve.