FOLLOW US

INSIDE THIS REPORT

• Defense lawyers are being urged to adopt AI responsibly to keep pace with prosecutors.
• AI can uncover critical evidence, but hallucinations and confidentiality failures remain serious risks.
• A federal ruling shows chatbot conversations may not be protected by privilege.

By Samuel López | USA Herald

The criminal courtroom is entering an AI arms race—and defense attorneys who refuse to learn the technology may risk leaving their clients at a growing disadvantage.

That is the central warning in a newly released National Association of Criminal Defense Lawyers report, Parity in Practice: The Defender’s Duty to Ethically Use AI. The report argues that as prosecutors’ offices increasingly experiment with enterprise-level artificial intelligence systems, defense lawyers have an ethical duty of competence to understand and responsibly use comparable tools.

The point is not to hand a criminal case over to a chatbot. It is the opposite: use technology to help a lawyer find what human time, staffing, and exhaustion can otherwise bury.

A properly governed AI system may help identify an inconsistency hidden across ten hours of body-camera video, locate a sentencing disparity, organize massive discovery productions, summarize lengthy records, or produce a first draft of a motion that an attorney then independently researches, verifies, and rewrites.

“AI is already inside the courtroom,” NACDL President Andrew Birrell said in a statement accompanying the report. “The defenders who learn to use AI well are the ones who are going to out-lawyer the other side.”

The report, authored by St. Mary’s School of Law Assistant Professor Mason Clark, frames the issue as one of parity. Prosecutors often work within more centralized systems, while public defenders and private criminal-defense lawyers may face smaller budgets, overloaded caseloads, and uneven access to technological support. If the state can use AI to move faster through discovery, develop litigation content, and identify patterns across records, the defense cannot simply ignore the same reality.

But the report’s call to action comes with a bright warning label: AI is an assistant, not a lawyer, investigator, expert witness, or confidential vault.

Generative AI systems can hallucinate—confidently producing false cases, quotations, facts, or legal analysis. A Stanford study cited in the report found significant error rates in legal AI products, including more than 17% incorrect information from Lexis+ AI and Ask Practical Law AI, while Westlaw’s AI-assisted research tool reportedly produced inaccurate responses at an even higher rate in the tested prompts.

That risk has already become painfully visible in courts nationwide, where lawyers have faced sanctions and reputational damage for submitting briefs containing fictitious authorities generated by chatbots.

Clark’s report separates potential AI use into three broad risk categories. Low-risk tasks include administrative drafting, internal policies, routine correspondence templates, training materials, and brainstorming general legal issues. Medium-risk work may include summarizing long documents, assisting with research, or preparing initial drafts of briefs and motions—but only when the attorney can thoroughly verify the output.

The highest-risk category involves confidential client information, case-specific factual analysis, strategic decisions, client counseling, and courtroom advocacy. In those areas, the report says, AI should support human judgment—not replace it.

That distinction is not academic. In February, U.S. District Judge Jed Rakoff ruled in United States v. Heppner that a criminal defendant’s unsupervised communications with a consumer AI chatbot were not protected by attorney-client privilege or the work-product doctrine. The ruling sent a direct message to litigants and lawyers: entering sensitive case facts or defense theories into a public-facing AI platform can create a discovery problem rather than a legal shield.

For that reason, the NACDL report advises defenders to carefully vet enterprise-level and legal-specific platforms with stronger privacy, security, and data-governance protections instead of casually uploading client information into personal consumer accounts.

The broader legal system is moving toward the same conclusion: AI is here, but it must be governed. The Judicial Conference’s rules committee has considered proposed Federal Rule of Evidence 707, which would require a reliability review for certain machine-generated evidence offered without a human expert. Congress is also considering the Research and Oversight of AI in Courts Act of 2026, a bipartisan proposal focused on AI speech-to-text and automatic speech-recognition tools in the judiciary.

For criminal-defense attorneys, the practical lesson is increasingly clear. A lawyer who uses AI without verification can harm a client. But a lawyer who refuses to understand AI while the prosecution deploys it responsibly may also fall behind.

The emerging duty is not blind adoption. It is disciplined adoption: protect confidentiality, verify every legal and factual claim, preserve independent judgment, and make sure technology strengthens—not silently compromises—the constitutional promise of a meaningful defense.

About the Author

Samuel López is a Senior Legal Analyst, investigative journalist, and legal researcher with more than two decades of experience analyzing litigation, and legal ethics.
👉 Follow us on X: https://x.com/RealUSAHerald