A Genuine Record Can Still Produce an Unreliable Conclusion

The proposed Rule 707 addresses the reliability of an AI-generated inference, while the ordinary authenticity rules ask whether the item offered in court is what the proponent says it is. This chapter keeps those questions separate and then examines the working proposal for generative-AI fabrication claims. The distinction gives the reader a practical test: a genuine recording can carry an unreliable analysis, and a reliable analytical system can be fed an item that is fabricated or misidentified.

That separation is easy to lose because the same digital exhibit may contain all the steps at once. A party may offer a video, an extraction, a generated report, and an AI conclusion about what the material means. The fact that the file is genuine does not establish that the conclusion is dependable. Conversely, evidence that a system is methodologically sound does not establish that this particular file came from the claimed source or was not altered.

An Authentic Item Is Not Necessarily a Reliable Inference

Authenticity and reliability answer different questions. Authenticity concerns identity and provenance: is this the recording, document, image, or computer output that the proponent claims it is? Reliability concerns method: did the process that produced the conclusion use adequate inputs, dependable principles, and a reliable application to the facts? The proposed Rule 707 belongs to the second inquiry when an AI output functions like expert testimony.

The official report makes the distinction explicit in its discussion of Rule 902(13), which provides a route for self-authenticating certain electronic evidence through certification. A certification may help establish that a computer system produced the offered output or that the record has the claimed origin. It does not assure that the information produced is reliable. The report says that self-authentication does not prevent an objection to the reliability of the information. In plain terms, a certificate can help answer “where did this item come from?” without answering “does the analysis inside it deserve trust?”

Consider a hypothetical system that analyzes a genuine surveillance recording. The recording may be properly preserved and accurately attributed to a camera. The system may still draw an unsupported conclusion because its training data do not fit the setting, its method was validated for a different task, or its application interprets an ambiguous image too aggressively. The item is authentic in the narrow sense while the inference remains unreliable. Rule 707 would address the latter problem under the proposed framework.

The reverse hypothetical is equally important. Suppose an analytical system has strong validation for a defined use, but the file supplied to it is a fabricated image or a recording attributed to the wrong person. The method may be reliable in the abstract, yet the conclusion cannot establish the fact the proponent claims because the input is not genuine. Methodological proof cannot repair a false premise. The authenticity challenge and the reliability challenge may involve the same exhibit, but they target different links.

This difference also explains why “self-authenticating” should not be heard as “self-proving.” Self-authentication can remove or simplify one foundational dispute.