The Fifth Circuit Court of Appeal in Louisiana has acquired two artificial intelligence servers, marking a significant step in its effort to develop in-house AI tools for managing legal workloads. Each of the Dell servers, equipped with Nvidia chips, cost approximately $65,000 at the time of purchase.
According to Judge Scott Schlegel and Chief Judge Susan Chehardy, the court’s approach is to experiment with open-weight AI models on its own infrastructure while also utilizing and testing commercial generative AI products. The court decided against committing entirely to either locally operated AI or commercial services, allowing it to compare different approaches as the technology and its costs develop.
Identifying Practical AI Applications
In collaboration with the Institute for the Advancement of Legal and Ethical AI, the project commenced with interviews of court employees to identify areas where AI could provide practical support. The initial focus is on developing tools for filing compliance and intake, which could potentially review proposed filings for issues such as excessive page lengths, incorrect pagination, and missing attachments before documents are submitted to the court.
The court also plans to explore the use of AI in document classification and routing. However, employees will remain responsible for verifying suggestions and conducting official filing reviews, rather than relying on AI to determine the course of a case. Future capabilities could include identifying nonexistent cases or inaccurate quotations in pleadings and detecting hidden instructions designed to manipulate an AI application reviewing a document.
Operating AI locally also creates its own governance requirements. The servers require additional power, cooling, maintenance and cybersecurity protections, The judges have acknowledged that housing technology on court premises does not automatically ensure its security. The experimental AI environment is being kept separate from the court’s existing case-management system. Initial applications will produce information for employees to evaluate rather than modifying official court records.
Related Post: Brazil Supreme Court deadlocks over elderly insurance case
This initiative appears to be the first publicly announced effort by a U.S. state court to purchase dedicated AI infrastructure for developing its own applications, although other state and federal courts are already experimenting with artificial intelligence. As courts across the U.S. consider how to incorporate AI into judicial work, the federal judiciary has cautioned against delegating core judicial functions, including decision-making and case adjudication, to AI.
Addressing Citation Errors and Accountability
The federal judiciary has emphasized that users remain accountable for work produced with AI assistance. One of the planned capabilities in Louisiana is particularly relevant to current challenges facing courts: identifying nonexistent legal authorities appearing in filings. Judges have encountered inaccurate citations and fabricated cases produced through AI-assisted legal drafting, increasing pressure on lawyers and courts to verify authorities before relying on them.
An automated citation check could provide another layer of scrutiny, although the Louisiana court says lawyers would remain responsible for their filings and court employees would continue to conduct independent reviews. The project could also help determine whether courts benefit from developing specialised AI applications around their own procedures rather than relying entirely on general-purpose commercial platforms.
Schlegel and Chehardy have not suggested that every court should purchase its own AI infrastructure. Instead, their approach is experimental, allowing the Fifth Circuit Court of Appeal to test costs, accuracy, and practical applications while developing internal technical expertise. If the experiment is successful, its significance could extend beyond the two servers, providing other courts with evidence about where AI can improve judicial administration and whether building court-specific systems offers advantages over relying solely on commercial technology.
