White & Case and Toshiba America deployed a closed AI platform during a six-week jury trial in what the firms believe is the first reported use of AI in a complex commercial trial. The platform, built by Eudia, was trained on the case's specific evidence, expert reports, and legal filings, enabling trial attorneys to access summaries and analysis from the courtroom in real time.
Trainers fed the system information from dozens of witnesses, hundreds of thousands of documents, and thousands of pages of testimony. The tool acted as "expert" counsel for urgent questions from trial lawyers while arguments were live.
Training the system on case specifics
The training series spanned three phases. Expert attorneys selected key case materials and provided markup to engineers. The broader trial team then gave daily feedback on the platform's outputs and user experience. Contract review attorneys finally tested the system against verified materials, flagging discrepancies to engineers.
"Once the senior team was satisfied with the quality of the outputs, the engineers essentially closed training for the platform to avoid negative effects from inaccurate feedback," the filing states.
Attorneys trained the AI on both breadth and depth, testing whether it could understand nuance and technical sophistication at the level of the trial team. The process was systematic and used human review at every stage to confirm accuracy.
Confidentiality, privilege, and verification safeguards
Due to the unsettled legal landscape around AI in litigation, the team implemented several protections. The platform used was a closed, professional-grade system custom-built for the case. Inputs and outputs were siloed, with access restricted to a short list of direct users and engineers working under attorney instructions.
"Attorneys and staff at every level of seniority were encouraged to be contrarians when judging AI outputs," the report states.
The platform linked responses back to underlying source documents so users could verify original materials. Training was limited to the specific litigation and could be deleted once the case concluded. The platform was never made available to witnesses or non-attorneys.
Use cases at trial included fuzzy searches
Trial teams used context-aware searches that did not require exact terminology, contrasting with traditional Boolean search systems that only match specific stored keywords. "We ran fuzzier searches, also run as context-aware searches," the firms reported.
Attorneys said the AI threat the ability to run semantically flexible queries allowed them to find relevant material without being constrained to specific legal jargon known beforehand.
Why this matters for legal professionals
Marshall trial AI can reduce the need for attorneys and staff who spend long hours on time-consuming tasks like document searching and low-level verification. For lawyers concerned about hallucinated case citations and privilege rulings, the Toshiba-White & Case case demonstration provides a concrete guide: closed systems, locked and separated training data, and multi-layer human verification can reduce risk. The model also shows that AI works best not as a replacement for judgment but as a tool that surfaces vetted evidence faster.
Legal professionals considering adding AI to their workflow should examine the AI Learning Path for Paralegals for skills that directly map to trial document review and contract analysis tasks. The AI for Legal collection covers additional implementation approaches for law firms and in-house teams.
Your membership also unlocks:
