Federal Judges Voice Optimism on AI, With Firm Limits on Oversight

Judges see AI as a way to enhance human effort

Everlaw Connect D.C. is a day-long thought leadership summit on shifts in the practice of law. The judicial panel included the Hon. Xavier Rodriguez (W.D. Tex.), the Hon. Allison Goddard (S.D. Cal.) and the Hon. Young Kim (N.D. Ill.). According to Everlaw's write-up, the judges stressed that AI tools can enhance human effort but cannot replace expert legal reasoning.

Judge Rodriguez, who uses generative AI in his chambers and teaches its applications at St. Mary's University School of Law, described the technology as a transformative force for democratizing the law by making resources more accessible. Judge Kim said AI can improve efficiency in privilege logs. He argued that AI-generated results may be superior to human reviews because they can offer improved clarity, precision, and less bias in choosing suggested withholdings.

The recap says legitimate concerns about misuse remain. The consensus, it states, is that as a regulated, well-supervised tool, AI can help legal professionals do more with less.

First drafts, hallucinations and confidentiality

Judge Goddard recommended a "first draft" approach to drafting, in which AI output is treated only as first-draft content. In her examples, AI synthesizes or rephrases existing information, such as summarizing parties' arguments or editing jury instructions for clarity. She emphasized that the "final draft" of any actual judicial reasoning, statutory construction, or constitutional adjudication must remain the product of independent human judgement.

On hallucinations, the panel took a proactive view. Rather than treating them as tool defects, the judges characterized them as a failure of attorney supervision.

On security, the panelists urged caution about sharing privileged information with consumer-grade AI platforms. Judge Kim suggested starting with tools that have no data retention and "no train" policies, so the models are not trained on privileged information. The recap also notes that under Rule 26 of the Federal Rules of Civil Procedure, the duty to maintain confidentiality remains fundamental.

Proposed Rule 707 and AI-generated evidence

The panel also discussed Proposed New Federal Rule of Evidence 707, a proposed amendment meant to help manage the admissibility of AI-generated evidence in federal courts. Under the proposal, AI-generated evidence offered without a human expert witness must meet the Daubert standard under Federal Rule of Evidence 702.

That standard requires the presiding judge to evaluate whether the methodology behind the evidence is scientifically valid and reliable. AI outputs would face the same scrutiny on peer review, error rates and general acceptance as traditional expert testimony. In Everlaw's words, Proposed Rule 707 seeks to apply the same strict reliability standards used for human expert witnesses to AI-generated outputs.

What it means for legal professionals

Everlaw concludes that the public sector legal community is entering an era defined by technical proficiency and growing optimism about AI. The practical advice from the panel is direct: treat AI outputs as first drafts, keep a strict duty of oversight, and choose platforms with strict no-train policies to protect client confidentiality.

The recap was written by David Pemberton, an associate content marketer at Everlaw, and reflects the company's summary of the event rather than a transcript. Its closing message is that independent human judgment and diligent supervision remain the foundation of the justice system.