AI-Generated Evidence Is Testing the Limits of Courtroom Standards Nationwide

In many states, courts are still evaluating evidence under a century-old legal standard that was never designed to handle artificial intelligence. As AI-generated evidence, from facial recognition to generative content, becomes more common in litigation, judges are increasingly being asked whether that old standard can even determine if such evidence is reliable.

The trend is the subject of a recent Bloomberg Law deep dive on AI-generated evidence in state courts, which quoted Beasley Firm attorney Andrew Warshawer, former Manhattan prosecutor and co-chair of the New York City Bar Association's subcommittee on Artificial Intelligence and Civil Liberties.

An Old Standard Meets a New Problem

The issue is showing up most often in criminal cases, where police and prosecutors increasingly rely on AI-driven tools, including facial recognition, gunshot detection, and license plate readers, to build cases. Defense attorneys say judges are largely declining to scrutinize the reliability of that technology.

Part of the problem is structural. A number of states, including Pennsylvania and New York, still use the Frye standard, a 1923 rule that admits scientific evidence if it has "general acceptance" in the relevant expert community. Most of the country, including the federal courts, moved decades ago to the more rigorous Daubert standard, which requires evidence to be tested and peer-reviewed before a judge lets it in. States that have stuck with Frye are now confronting whether a century-old rule can meaningfully evaluate AI-driven tools at all.

Warshawer put the core issue plainly:

"The bench is 'woefully unprepared' to deal with this technology... We don't have all the answers for this. It's an emerging technology."

Why the Usual Legal Analogies Break Down

Courts typically evaluate new technology by analogy. An email server, for instance, gets treated like a filing cabinet. Warshawer argues that approach doesn't hold up for AI-based tools, because the reasoning behind an AI system's output often isn't knowable, even to the vendor:

"Nobody knows how these things freaking work, and the companies that make them have no interest in telling you."

That opacity puts judges in a difficult spot, asked to gatekeep evidence generated by systems that resist the kind of transparency the Frye and Daubert standards were built around.

What a Real Fix Would Take

Warshawer isn't optimistic about a quick legislative solution. In his view, closing the gap would require bar associations, courts, prosecutors, and both the plaintiffs' and defense bars to negotiate a comprehensive set of AI evidentiary rules, a process he estimates would take 18 to 24 months even under ideal conditions.

"That's what I think is ideal. I just think in New York, and probably in most of the country, it's unachievable."

The Takeaway

Until courts, or legislatures, catch up, litigators on both sides are left arguing AI evidence questions case by case, with judges applying decades-old standards to technology that didn't exist when those standards were written. For practitioners in Pennsylvania and other Frye jurisdictions, this isn't a distant New York issue. It's a gap worth watching closely for anyone litigating cases where AI-derived evidence, from surveillance tools to generative content, is in play.

Read the full Bloomberg Law article, "AI Evidence Raises Challenges for New York Courts, Lawyers Say," for the complete reporting.