Amid the rise of generative artificial intelligence, poised to reshape society and propel humanity to unprecedented heights, fears persist that AI also wields the power to upend human civilization entirely. This contradiction has fueled an ongoing discourse across many disciplines aimed at maximizing AI's benefits while mitigating its potential harms and risks. The legal profession is no exception, as the rise of ChatGPT and other generative AI tools highlights AI's transformative potential for the legal system.
While AI's potential is significant, it also poses risks to the fundamental elements of the legal system. Prominent AI experts caution against a possible crisis of widespread misinformation, which may leave individuals unable to distinguish "what is true." These fears are not merely theoretical, as AI deepfakes have already reached a level of sophistication that blurs the lines between reality and fabrication in images, videos, and voices. Considering that deepfake fraud increased by 3,000% in 2023 compared to the prior year, courts will undoubtedly face mounting challenges with verifying digital evidence. See "Deepfake fraud attempts are up 3000% in 2023 — here's why," The Next Web, November 2023, https://thenextweb.com/news/deepfake-fraud-rise amid-cheap-generative-ai-boom.
After some lawyers made headlines for submitting legal briefs with fictitious case citations that were generated by AI, several judges have updated their standing orders to address the use of AI-generated content in court f ilings.However, as AI-generated evidence becomes more prevalent and increasingly indistinguishable from its non-AI counterparts, courts will also need to address complex issues concerning the authenticity, reliability, and admissibility of trial evidence generated or enhanced by AI.
AI-generated evidence not only complicates the judicial process but also threatens to undermine its integrity. The integration of AI into the legal and judicial systems raises complex ethical challenges and profound implications for lawyers and the judicial landscape. This article discusses the multifaceted challenges presented by AI-generated evidence, outlines the principles governing its admissibility, and discusses what is currently being proposed to address the introduction of AI-generated evidence in courts.
What is AI-Generated Evidence?
AI-generated evidence can include a variety of materials, such as documents, photos, videos, and communications created by AI, as well as data analyses and patterns identified through machine learning algorithms. Additionally, AI software can enhance the quality of existing audio, video, or images. This enhancement involves adding elements that the AI determines should be present rather than just modifying what already exists. For example, AI might fill in missing pixels on an image or improve the clarity of a video (thus altering the original) instead of merely manipulating the original by zooming in, speeding up, or slowing down footage.
Of particular concern is the risk of AI being used to manipulate videos and images to create "deepfakes" — i.e., artificial images, video clips, and audio recordings created by AI that are fake but appear to be real.
The Challenges of AI-Generated Evidence
Unlike traditional evidence, which people can often track and verify through direct sources, AI-generated content might lack a clear source or come from complex processes that are hard to audit. This raises concerns about the integrity of the data used in AI systems, the accuracy of the algorithms, and the possibility of bias or errors in their results. AI systems are only as unbiased as the data on which they are trained. Historical data can carry biases, and AI can perpetuate or even intensify these biases.
Furthermore, proving the authenticity of and maintaining a chain of custody for AI-generated evidence is challenging. The opaque nature of AI processes complicates the parties' ability to prove the integrity of evidence that AI has generated.
Compounding matters is the ease with which even authentic digital evidence can be challenged by suggesting that AI may have surreptitiously altered it, raising ethical concerns for attorneys. Courts have encountered challenges in admitting evidence because parties argue that any digital content could be a deepfake. This "deepfake defense" alleges that audio or visual evidence presented at trial is fabricated, seeking to exploit skepticism among judges or jurors. See The Other Side Says Your Evidence Is A Deepfake. Now What?, Law360 (Dec. 21, 2022), https://www.wilmerhale.com/en/insights/publications/20221221-the-other-side-says your-evidence-is-a-deepfake-now-what .
Addressing these issues, along with the complex nature of many AI systems, almost certainly requires expert testimony to clarify how the evidence was generated and maintained. Considering such evidence will likely lead to a time-consuming trial within a trial about the non-peer-reviewable process used by the AI model.
Principles for Addressing the Admissibility of AI-Generated Evidence
The rules of evidence, specifically Rules 401, 402, and 403 of the Ohio Rules of Evidence, give trial judges the gatekeeping responsibility of determining the admissibility of evidence, including AI-generated evidence. To establish the relevance of AI evidence, the party presenting must demonstrate how the AI system works (i.e., how it generated its outcome) and how the evidence will aid, rather than confuse, the jury in reaching a just verdict. This requires disclosing enough information about the AI system's training data, development, and operational mechanisms so that both the opposing party and the judge can evaluate it effectively. Several factors influence the admissibility and relevance of AI evidence. While the accuracy and reliability of AI systems are paramount, the interpretability of complex algorithms and the potential for privacy violations due to extensive data use are also significant considerations. Additionally, the timeliness of AI-generated evidence and the inherent biases in AI systems should greatly affect their relevance and, in turn, their admissibility in court.
The authentication of evidence, which is outlined in Rule 901 of the Ohio Rules of Evidence, requires parties to verify that the evidence is what it claims to be, including AI generated evidence. The rules most relevant to the authentication of AI-generated evidence are 901(b)(1) and 901(b)(9), which pertain to witness testimony and evidence that describes a process or system producing accurate results, respectively. Challenges in authenticating AI evidence again include the opacity of AI algorithms, potential biases in training data, the quality of the data used, compliance with regulatory standards, and a general lack of legal expertise with AI technology. These factors can complicate the authentication process and raise questions about the reliability and accuracy of the evidence.
However, the possibility that deepfake media may be submitted as unadulterated evidence does not give litigants carte blanche to baselessly question evidence either. Doing so may violate professional rules against making frivolous arguments, baselessly denying factual contentions, or engaging in harassing, delaying, or costly motion practice.
What's Next for Rules Regarding AI-Generated Evidence in the Courts?
In late 2024, theU.S. Courts Advisory Committee on Evidence Rulesproposed amendments that seek to govern evidence generated by AI technology. The Committee has proposed three potential amendments:
- Create a new Fed. R. Evid.707 that explicitly subjects AI-generated outputs to the same standards as a human expert witness as required by Rule 702 (a) – (d).
- Add an additional Section (c) to Fed. R. Evid.901, which standardizes a test for challenging evidence purportedly created by AI.
- Amend Fed. R. Evid.901(b)(9) to include examples of evidence that satisfies Fed. R. Evid.901(a) requirements relevant to artificial intelligence.
U.S. District Judge Edmund Sargus, who serves in the Southern District of Ohio in Columbus and is a member of the Advisory Committee on Evidence Rules, explained the proposal as requiring that "AI-generated opinions meet theDaubert standards" that have traditionally applied to expert witnesses' testimony. Judge Sargus further stated he expects that only one of the proposed rule changes will ultimately be adopted: either one of the amendments to Fed. R. Evid. 901 or the addition of Fed. R. Evid. 707. In the interim, Judge Sargus suggested that lawyers be prepared for judicial review of evidence created through AI technology under a Daubert test. See Proposed Evidence Rules Tackle AI Evidence, OBLIC, January 21, 2025, https://www.oblic.com/resources/oblic news/01/21/2025/proposed-rules-ai-evidence/.
While similar amendments to the Ohio Rules of Evidence have not yet been proposed, one thing is for sure: AI is moving faster than ever, and sooner or later (if not already), all courts will have to grapple with the foundational elements of how AI is treated differently from humans and how it impacts existing laws. A comprehensive approach is crucial for addressing challenges related to AI-generated evidence in the legal system. This includes establishing clear standards and regulations to ensure reliability and fairness. Furthermore, lawyers need to understand AI's capabilities and limitations in order to use and challenge AI-generated evidence more effectively and ethically.
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.
