Products
SCRS — the AI data firewall Other Me — the workspace Compare both products
Solutions
All solutions AI Assistants Email + Projects CRM Live Chat Ticketing Flows HR-lite Control Center
By industry
Accountancy Legal Mortgage & IFA Healthcare
Company
Custom Solutions Pricing Blog About Us
Log in
SCRS login Other Me — Business Other Me — Family Get started with SCRS
Architecture · · 5 min read

Per-Matter AI Vaults Explained: How Chinese Walls Work in AI

AS

Founder & CEO, Pop Hasta Labs

From my perspective, Chinese walls are one of those concepts that every solicitor knows exists in principle and few firms can demonstrate structurally. In the paper-file era, Chinese walls lived in the filing cabinet and the matter inception process. In the AI era, they need to live in the software — or they don’t live at all.

What a per-matter vault is

A per-matter vault is an encrypted storage area dedicated to a specific client engagement. When you create Matter 2036-A4, a new vault is created; all documents, correspondence, AI interactions related to that matter go into it. Matter 2036-B7 has its own vault. The two vaults don’t know about each other.

When the AI works on Matter 2036-A4, it can only retrieve from Matter A4’s vault. The software enforces this at the retrieval layer — the AI literally doesn’t have permission to query other vaults during that session. Specially, this isn’t a prompt instruction (“please only use Matter A4 data”); it’s a technical boundary the AI cannot cross.

Why this matters structurally

Prompt-based isolation is defeatable. If the AI is smart enough, it can reason around instructions; if the user is determined, they can craft prompts that coax the AI to ignore the isolation. Retrieval-layer isolation is not defeatable by prompting. The AI cannot retrieve what the software doesn’t expose.

For SRA-regulated firms, this is the difference between Chinese walls in a policy document (which the SRA has long treated as weaker) and Chinese walls in the architecture (which genuinely close the risk). Apart from this, for litigation firms handling matters that are formally adverse, the architectural guarantee is what lets the firm work on both sides without raising professional conduct issues — provided the fee-earners are properly separated, the AI layer maintains the separation automatically.

Learn more

Full law-firm workflow on the Law Firms solution page. Related: AI bundle review preserving privilege, SRA guidance on AI for solicitors. SCRS is self-serve — get started whenever you are ready.

AS

Abhishek Sharma

Founder & CEO of Pop Hasta Labs. Building SCRS — the patent-pending AI data firewall — and Other Me, the governed workspace built on it. Based in London.

Govern the AI your team already uses.

SCRS keeps client data out of the model — and Other Me is the whole governed workspace, built on it. No sales calls; set yourself up in minutes.