Last week’s Everlaw Summit had several amazing sessions. Here are six takeaways from the Everlaw judges panel at last week’s conference.

The session “The Judicial Perspective on AI and Legal Technology” at the Everlaw Summit was moderated by Gloria Lee, Everlaw’s Chief Legal Officer. The panel featured Chief U.S. Magistrate Judge Willie Epps Jr. of the Western District of Missouri; U.S. Magistrate Judge Allison Goddard of the Southern District of California; and retired Arizona Court of Appeals Judge Samuel A. Thumma, Director of the Arizona Supreme Court Center on Forensics and Artificial Intelligence.

Their discussion offered six takeaways about AI’s opportunities, risks, and limits. They also referenced the publication the Judicious Judge’s Guide to Generative Artificial Intelligence and Large Language Models (which we covered here) that Judge Goddard and Judge Thumma participated in writing. So, here are six takeaways from the Everlaw judges panel.

Prestigious Law Firms Are Not Exempt from Judicial Skepticism

Judge Epps described heightened scrutiny of filings, including those submitted by prominent firms. A firm’s reputation cannot substitute for verification when attorneys use generative AI without adequate oversight. Fabricated citations, inaccurate descriptions of holdings, and invented quotations can reach the court regardless of who signs the brief.

For judges, the concern extends beyond correcting attorneys’ mistakes: an inaccurate filing can contaminate a judicial opinion and undermine confidence in the court.

“I just don’t want to be a judge you read about who let something out of chambers that is just damn wrong,” Judge Epps said. “That would be incredibly embarrassing. You all have seen those headlines, and I don’t want to be that guy.”

Lawyers must not only verify their submissions, but judges must also independently scrutinize the materials on which they rely.

Cybersecurity Challenges Are Sending Some Paperless Courts Back to Paper

Digital modernization doesn’t always move in one direction. Judge Epps described how cybersecurity threats have forced changes in courts that previously embraced paperless administration.

Attacks targeting electronic filing systems and concerns about unauthorized access to sealed records have prompted some courts to require physical delivery of sensitive filings, meaning that records once accessible electronically may now require clerk’s office staff to deliver paper files to chambers.

Convenience and efficiency depend on trustworthy infrastructure. When confidentiality is threatened, courts may be forced to accept slower processes to protect sensitive information.

Prompt Injection Raises Risks Beyond Misleading a Judge

The panel discussed prompt injection within filings, referencing examples including this case in Brazil and this one in Connecticut: in both cases, hidden white text – invisible to a person reading a brief but accessible to a language model processing it – was used to try to sway AI models.

Judge Thumma characterized the tactic as “about as dirty as it gets.” Judge Goddard distinguished between attempts to influence legal analysis and attacks on connected AI agents. The judges agreed that human review and adversarial scrutiny provide safeguards against hidden instruction determining a ruling, while an agent with access to email, credentials, or other systems presents different risks, including unauthorized disclosure of information.

The discussion also resisted treating self-represented litigants primarily as a security concern. Judge Thumma emphasized AI’s potential to improve access to justice, including helping people seeking emergency protection navigate complicated intake forms during a crisis. Courts must balance malicious uses with beneficial assistance.

Broad AI Disclosure Rules Can Create More Confusion Than Clarity

Nationwide, courts have issued hundreds of local rules or standing orders regarding GenAI disclosure with an estimated 884 distinct orders across federal and state systems (note: I found the source that they referenced and it’s up to 908 orders now).

The panel criticized proliferating standing orders and local rules requiring AI disclosures or certifications when existing procedural and ethical rules – such as Federal Rule of Civil Procedure 11, the duty of competence, and professional conduct rules – already obligate attorneys to verify the truth and accuracy of every submission.

“These standing orders are not necessary”, said Judge Goddard. “I saw a standing order that was issued in the last three months that says you cannot use artificial intelligence to draft any brief. It doesn’t say generative AI. It says artificial intelligence.” As she noted, those general restrictions extend to even grammar checkers like Grammarly.

Judge Thumma described a more constructive approach in Arizona: surveying judges across court levels to understand actual AI use, technical knowledge, and training needs.

AI Can Save Significant Time Without Replacing Judicial Judgment

Judge Goddard described using AI to assist with Social Security disability appeals by preparing factual summaries and draft structures informed by her previous opinions. She reported it reduces the time it takes to draft orders up to 75%!

She also described collaborative work involving judges, law clerks, technology personnel, administrators, and practitioners to develop useful prompts and workflows, including generating routine scheduling orders.

Judge Epps envisioned a secure AI system functioning as a “third law clerk,” drawing on his previous rulings to help prepare initial drafts. Such assistance could be especially valuable in state courts, where many judges manage substantial caseloads without law clerks.

Judge Thumma also highlighted the value of court data analysis. Differences in eviction hearing attendance by day of the week showed how scheduling changes could help more families appear and potentially avoid default judgments.

Still, the panel drew firm boundaries. AI may organize records, summarize facts, and assist with drafting. Evaluating credibility, weighing conflicting evidence, exercising discretion, and making binding determinations remain human judicial responsibilities.

Professional Courtesy Still Matters When AI Errors Surface

Judge Epps urged lawyers to exercise “grace” when opposing counsel submits inaccurate authorities or AI-generated errors. Rather than immediately seeking sanctions, attorneys should consider contacting counsel and providing an opportunity to correct the mistake.

Still, courtesy doesn’t eliminate accountability. The panel discussed a case in the Fifth Circuit involving a judicial order containing serious inaccuracies, including invented quotations. The appellate court vacated the order, remanded the case, and reassigned it to another judge.

Bottom Line

AI can support faster, more accessible court operations, but responsibility can’t be delegated to software. Lawyers and judges still must verify the work, protect confidential information, and preserve independent human judgment to maintain trust. In other words, same as it ever was. 😉

So, what do you think of the six takeaways from the Everlaw judges panel? Please share any comments you might have or if you’d like to know more about a particular topic.

Image created using ChatGPT, using the term “panel with a robot moderator and three robot judges as panelists”.

Full Disclosure: AI assisted with this write up. Ironic, isn’t it? 😉

Disclaimer: The views represented herein are exclusively the views of the author, and do not necessarily represent the views held by my employer, my partners or my clients. eDiscovery Today is made available solely for educational purposes to provide general information about general eDiscovery principles and not to provide specific legal advice applicable to any particular circumstance. eDiscovery Today should not be used as a substitute for competent legal advice from a lawyer you have retained and who has agreed to represent you.