Air-gapped AI in practice
An illustrative example based on a typical engagement: how sensitive evidence review at a UK law firm moves from slow, cloud-exposed manual work to a controlled, fully offline AI workflow.
UK regional law firm — litigation & regulatory practice
The client handles litigation and regulatory investigations involving highly sensitive personal and commercial data. Their compliance team needed AI-assisted document review but could not send case material to cloud AI services without breaching client confidentiality obligations and UK GDPR commitments. This is an illustrative scenario showing a typical engagement profile, not a named client case study.
The challenge
Each investigation produced encrypted drives containing hundreds of thousands of documents. Reviewing them manually took analysts most of a working day per drive, and the firm's clients were asking hard questions about where their data went when AI tools were involved.
The firm's compliance director set three non-negotiables: case data must never leave a controlled environment, the original evidence must remain forensically intact, and the whole process must be auditable end to end.
The engagement, step by step
Drive intake
The firm couriered an encrypted evidence drive to our facility. The drive was logged, photographed, and connected only through a forensic write-blocker, preserving the original media bit-for-bit.
Air-gapped processing
A copy of the dataset was analysed on a physically isolated workstation running local AI models. No network interfaces, no wireless hardware, no cloud connection of any kind.
Analyst-verified output
AI-generated summaries, entity extraction, and chronologies were reviewed by a human analyst before delivery. Only the verified report — never the underlying data — left the room on encrypted media.
Zero-retention close-out
Working copies were securely erased at the end of the engagement and the original drive returned. The firm received a certificate of erasure alongside the final report.
Before and after
Time to review one drive
Before: 6–8 hours of manual keyword searching and reading
After: Under 90 minutes, including analyst verification
Data exposure
Before: Files copied to cloud AI tools outside the firm's control
After: Nothing leaves the air-gapped workstation — ever
Audit trail
Before: Ad-hoc notes; difficult to evidence to regulators
After: Full chain of custody and zero-retention record per session
Cost per matter
Before: External e-discovery billed per gigabyte
After: Fixed engagement fee; predictable per-matter cost
Figures are illustrative of a typical engagement of this kind, not measured results from a named client. Actual results vary by dataset size, encryption, and review scope.
"For the first time we could tell a client, in writing, that their data never touched the internet. That answer alone has changed how we pitch for sensitive work."
Facing the same problem?
If your team handles evidence, investigations, or regulated data that cannot go near the cloud, tell us about your drives and we'll scope an engagement.