The Williamsburg-based National Center for State Courts, working with the Thomson Reuters Institute, recently released a report on how artificial intelligence is reshaping the authentication of digital evidence in state courts. The partnership published “Evidentiary Issues Raised by Artificial Intelligence” to help judges and court leaders navigate the growing use of AI-generated and deepfake material. The report reflects NCSC’s Strategic Agenda, which calls for educating courts on AI’s impact, and responds to a widening gap between tools built to fabricate convincing evidence and those built to detect it, the partnership said.
The report noted that two broad strategies exist for courts confronting AI-generated evidence. Court systems can expand access to technical experts and detection technologies, though such tools should complement rather than replace legal frameworks for evaluating evidence. Updating rules and procedures is another option, as Utah has done by requiring machine-generated evidence offered without expert testimony to meet the same reliability standards as expert witnesses. Connecticut, Florida, Louisiana, New York and Oklahoma have introduced similar rules addressing artificial intelligence in legal proceedings, the report added.
The document offers judges practical guidance, including questions to ask about the source and authenticity of digital evidence presented in court, the report said. Recent rulings show the stakes: a California court dismissed a civil case after finding plaintiffs had submitted deepfake videos and altered images, and a New York court reversed a family court abuse finding because video evidence had not been properly authenticated, it said. That guidance supports NCSC’s Strategic Agenda, which calls for recognizing artificial intelligence’s impact on court operations and offering advice on its ethical and practical use. That education push extends beyond the report itself, from in-person workshops to NCSC’s eCourts technology conference, the organization said.
The challenge extends beyond typical forgeries. Courts have long relied on chain-of-custody requirements, expert testimony and authentication procedures to evaluate evidence, and those tools still apply to some degree. But the report noted that deepfakes differ in scale and speed and that artificial intelligence has made sophisticated fakes accessible to people with little technical skill. This content is often convincing enough to fool trained observers, and detection tools have struggled to keep pace with AI's ability to evade them. The report asserts that these factors, including the mere existence of AI technology, have begun eroding trust in the authenticity of digital evidence overall.
Detection methods generally fall into two categories, according to the report. Inference-based detection examines the media itself for signs of manipulation, such as inconsistent lighting, unnatural facial expressions or artifacts specific to AI-generated content. Provenance-based detection instead looks at a file’s origin and history, including metadata, timestamps, geolocation data and digital signatures. Even with these tools, automated detection often loses accuracy when confronted with new AI models it was not trained to recognize.
Updating court rules can take two forms, NCSC said. Some experts argue that courts can address AI-generated evidence under existing mechanisms, such as expert testimony and authentication procedures, the same tools used for prior technological shifts. Others have pursued specific rule changes, including Connecticut's compliance rule for generative AI, Florida's rule on signatures and representations to the court, Louisiana's attorney code provision, New York's rule on AI technology use, and Oklahoma's rule on generative AI in criminal appeals. These rules generally place the burden of verification on the party submitting evidence, and some allow sanctions for inaccuracies, according to the partnership.
Beyond policy recommendations, NCSC has outlined steps individual judges can take, it said. For acknowledged AI-generated evidence, judges can issue pretrial orders requiring early disclosure of any intent to present or challenge such material. When parties disagree on whether evidence is AI-generated, judges are encouraged to ask the proffering party about the evidence's source, who has had access to it, what steps were taken to prevent tampering, and whether other data could help confirm its authenticity, it added.
NCSC has also taken its AI education efforts directly into courthouses, the organization said. It recently partnered with the Dallas County, Texas, Clerk's Office to deliver an in-person AI Readiness & Literacy Workshop for leadership and key stakeholders, covering AI fundamentals, governance and policy considerations, and opportunities for responsible AI adoption in court operations. Participants also used AI tools during the session to draft governance documents and policy language in real time. The workshop helped the office begin shaping its own AI policy, the Dallas County clerk said.
That training also scales up to a national conference NCSC hosts, eCourts, which will be held in Las Vegas later this year. The meeting is touted as bringing together judicial and IT leaders to examine how emerging technologies, including artificial intelligence, are reshaping court operations and the administration of justice. The event will include new workshops and special interest group discussions, the event's organizers said.