For in-house legal teams that need AI document review with strict data residency, the practical path is a dedicated intake pipeline running on an isolated model deployment, not a shared chatbot. AT&T's LegalEdge, built with OpenAI, is the clearest public template for that pattern right now.
According to OpenAI, AT&T's legal team uses LegalEdge to rethink how matters are intaken and routed, so attorneys spend time on judgment calls instead of triage. The stated goal is to make legal knowledge easier to access across the business and scale the team's expertise without hiring more lawyers for repetitive work.
What a small legal team can take from this
You do not need AT&T's budget to copy the shape. A thin intake agent that classifies requests, pulls relevant clauses and precedents, drafts a first answer and escalates edge cases covers most day-to-day volume - architecturally close to the AI legal assistant I built for contract archive audits. Keep documents in your own tenant or an in-country cloud, log every retrieval, and let the LLM produce only the draft - the lawyer still signs.
Why it matters: The real win in LegalEdge is not the model, it is the workflow boundary: humans decide, software handles intake and lookup. A small firm can replicate that pattern this quarter with Claude, OpenAI or an open model running behind its own perimeter.
Source: OpenAI