What happens to privileged data inside the AI tools most attorneys are already using

When a client walks into your office and hands you their most sensitive documents, they are not just hiring a lawyer. They are extending a level of trust that carries legal weight, ethical obligations, and — if you are the attorney who loses a case because privileged material was compromised — professional consequences that do not go away.

Most attorneys understand this on an abstract level. Far fewer have asked the next question: where does that data actually go when you upload it to the AI tools you use every day?

“The contract says ‘we don’t train on your data.’ The architecture tells a different story.”

The Vendor Agreement You Probably Did Not Read in Full

Every major cloud-based legal AI platform — the research tools, the document summarizers, the drafting assistants — routes your documents through third-party infrastructure to run their AI. That means your client’s deposition summary, the financial records in your product liability case, the medical history in your personal injury file, all leave your environment and travel through compute infrastructure that you do not control.

Some platforms say they do not use your data for training. That is a contractual promise, and contracts get amended. Some say they anonymize data. Anonymization has an imperfect track record in any industry that has tried it at scale.

None of them can say: the AI inference ran inside your environment. Because for cloud platforms, it cannot. That is a structural limitation of how they are built, not a policy decision they can change with an update.

Why This Matters More in 2026 Than It Did in 2023

Institutional clients — the corporations, healthcare systems, and financial firms that retain boutique litigation counsel for their highest-stakes matters — are beginning to audit this. Procurement teams are asking vendors to document their AI data flows. Risk officers are reviewing which AI tools law firm partners are using. Some engagement letters are now including AI usage clauses.

The solo and small firm attorneys who cannot answer the question “where does your AI inference run?” are going to find themselves on the wrong side of these conversations. Not because their work is inadequate, but because their architecture is.

The Architectural Answer

Lex Arca was built from the ground up on a local-first, BYOS (Bring Your Own Storage) architecture. Your documents live in your vault — built on Cloudflare R2 and Supabase infrastructure that belongs to your firm. The AI inference runs inside that environment.

This is not a marketing claim. It is the reason Lex Arca can produce an AI Compliance Certification and a Vendor Risk Audit Report — documents that pull live data from your case vault and generate a documented, verifiable record of exactly how your AI operates. No cloud-based competitor can produce equivalent documentation because their architecture does not support it.

When a client asks how you protect their data, “we have a local-first private vault with append-only, tamper-evident audit trails and documented AI compliance certification” is a different answer than “we use a reputable cloud platform.”

“Your clients are beginning to ask which one you can give them.”

The Compliance Layer Is Already Built

Phase 1 compliance features are live inside the Lex Arca platform. From within any active case view, you can generate an AI Compliance Certification with your case’s live UUID and SHA-256 ledger hash. From the firm-level Settings tab, you can generate a full Vendor Risk Audit Report with an A+ risk score documentation.

These are not marketing collateral. They are functional compliance documents tied to your actual vault architecture — the kind that holds up in a client conversation, a bar inquiry, or a procurement review.

The attorneys who lead on this will not just protect themselves. They will win client relationships that require the documentation to even begin.

→  Trial cases are preloaded. Just show up. vault.lex-arca.com