Our Take
Wilson has identified a real constraint (client data isolation in shared legal AI) that larger platforms ignore because the fractional GC market is too small to chase, but the architecture claim rests entirely on the vendor's word.
Why it matters
Fractional general counsels juggle confidential information for multiple companies simultaneously, making data leaks a bar-complaint risk. Most legal AI platforms pool client data on shared infrastructure, which is a disqualifier for this use case but invisible to solo practitioners.
Do this week
Fractional GCs: audit your current AI tools (Harvey, CoCounsel, Legora) against your bar association's data-handling rules before uploading client documents, then contact Sapphire for a private-tenant demo if you manage three or more concurrent clients.
Sapphire Legal launches with two paying customers
Brett Wilson, a former VMware enterprise IT veteran laid off after Broadcom's acquisition, bootstrapped Sapphire Legal and now operates with two customers. One is Andrew Friedman, a U.S. lawyer based in Portugal who manages banking, fintech, and payments clients. Wilson's premise is structural: every major legal AI platform (Harvey, CoCounsel, Legora) operates on shared infrastructure, meaning client data trains or touches models that other firms' data also touches.
For a fractional general counsel managing five client matters simultaneously, Wilson argues this is not a nuance but a disqualifying architecture. Sapphire's alternative is a private-tenant model. When an FGC onboards a client, the system spins up a separate database, storage bucket, encryption key, and search index for that client. Wilson describes it as a true architectural wall, not simply row-level security or access permissions.
The platform functions as a full legal operating system. It includes document intelligence (generation, risk assessment, contract benchmarking, anomaly detection, summarization), client intelligence (case posture evaluation and success prediction against a 12-million-case index), call analysis with sentiment detection, and validated citations tied to an integrated Westlaw license or internal index to mitigate hallucinations. It also covers practice-management basics: contacts, time tracking, billing, client portal, and e-signatures.
Pricing in the U.S. is $499 per month for the command center plus $999 per month per managed customer (Sapphire says FGCs can pass this through to clients as a line item). In the U.K., it is £499 per month for the base plus £250 per additional customer.
The fractional GC market is invisible to venture-scale legal AI
The market size is murky. Wilson estimates 50,000 fractional GCs in the U.S. and U.K., but he acknowledges no definitive data exists. ChatGPT's research suggested a reasonable range of 5,000–10,000 core fractional GCs in the U.S. and 10,000–25,000 globally, with lower confidence because the category is not separately tracked.
Whatever the true number, it is too small for Harvey and Legora to prioritize but too sophisticated for consumer legal tech. That gap is where Wilson is betting. His product is purpose-built for context switching and client isolation rather than competing on drafting or research speed.
Friedman, the first customer, reported recovering about 10% of his working time in three weeks and felt confident enough to take on another client. He moved from spiral notebooks, Google Drive folders, and email inboxes to a consolidated platform. The appeal, he said, is that fractional GCs are treated as the main event rather than as a small fish in a big pond.
Verify the isolation claim before adopting
Wilson spent months parsing 12 million cases and documents from Court Listener and configured automated systems to pull updated cases and statutes from all 50 states. He runs his own inference engine rather than routing everything through a third party, which he says keeps costs and data exposure in check. The platform learns each client's voice and context as the attorney uploads documents and reviews matters, allowing the AI to adopt the client's distinct style.
The private-tenant model is the core differentiation. Before adoption, fractional GCs should request a technical audit of the isolation architecture, not just marketing claims. Ask whether the separation is enforced at the infrastructure layer (separate databases, encryption keys, compute) or at the application layer (access controls, row-level security). The difference determines whether a misconfiguration or insider threat can leak one client's data to another.
If isolation is genuine and verifiable, the value proposition is clear: a lawyer managing five competing clients can use the same platform without the bar-complaint risk of shared-infrastructure tools. If it rests on application-layer controls alone, the risk profile remains unchanged.