A digital cognitive assessment device clears FDA review, ships to primary care clinics, and starts producing scores. Two years later, one of those scores lands in a personal injury file, and an attorney asks a forensic neuropsychologist whether it proves their client has a brain injury.

The answer is almost always no, and that has little to do with whether the device works. FDA clearance and courtroom admissibility test different things, and the gap between them catches device developers, clinicians, and litigators off guard. For anyone building cognitive assessment technology, that gap is a product problem, and it is cheaper to find than a deposition is.

What FDA Clearance Actually Establishes

Computerized cognitive assessment tools sit under 21 CFR 882.1470. The regulation defines the device as a prescription device that “uses an individual’s score(s) on a battery of cognitive tasks to provide an interpretation of the current level of cognitive function.” These are Class II devices with special controls. Since a 2019 amendment, they are exempt from 510(k) premarket notification unless the device is intended for diagnostic assessment of specific diseases or conditions and relies on visual cues, auditory cues, or functional use of the hand.

The regulation then draws a line most product teams underweight. The device “is used only as an assessment aid to determine level of cognitive functioning for which there exists other valid methods of cognitive assessment and does not identify the presence or absence of clinical diagnoses.” It “is not intended as a stand-alone or adjunctive diagnostic device.” Special controls require labeling that carries both warnings. The classification opened with a De Novo request in 2013 and was finalized in August 2015.

The regulation rules out the exact role attorneys most often want the device to play, which is corroborating a diagnosis someone else already made. Clearance establishes that the device performed acceptably against its special controls for its labeled intended use. It does not establish that the device can identify why a person’s cognition changed, when it changed, or whether the person put forth real effort during the test. Every one of those questions is what a personal injury or disability case turns on.

What a Court Asks Instead

Under Daubert v. Merrell Dow Pharmaceuticals, a judge screening scientific testimony looks at whether the technique can be and has been tested, whether it has been peer reviewed and published, its known or potential error rate, the existence of standards controlling its operation, and its acceptance in the relevant scientific community. The Court called the inquiry “a flexible one” and declined to “set out a definitive checklist,” so judges weigh these differently case to case.

Federal Rule of Evidence 702 was amended effective December 1, 2023, and the change matters here. The rule now requires the party offering the expert to demonstrate “to the court that it is more likely than not” that each admissibility element is met, and subsection (d) now asks whether the expert’s opinion “reflects a reliable application of the principles and methods to the facts of the case.” Courts had been treating shaky application as a weight issue for the jury. The amendment pushed it back to the judge.

Florida runs the same test. The Legislature enacted Daubert amendments to section 90.702 of the Florida Evidence Code in 2013, and the Florida Supreme Court adopted them as procedural rules on May 23, 2019, receding from its earlier decision to keep the Frye standard. In our practice across Florida’s judicial circuits, the effect is visible: the methodology behind a number gets examined, not just the number.

Take note of what is absent from that list: No Daubert factor asks whether a device holds FDA clearance. Clearance can support the reliability argument. It does not make it.

6 Places Digital Cognitive Tools Break Down Under Cross-Examination

1. Norms that do not match the examinee: A normative sample drawn from English-speaking, college-educated volunteers produces a misleading percentile for a 58-year-old Spanish-dominant warehouse worker with 8 years of formal education. Opposing counsel will ask who was in the reference group. If the demographic breakdown is not published, the score is hard to defend.
2. No performance validity measure: The American Academy of Clinical Neuropsychology’s 2021 consensus statement on validity assessment takes the position that the validity of each test result should be assessed in all clinical and forensic examinees. Most screening-oriented digital tools produce a cognitive score with no embedded check on whether the person was engaged, distracted, or deliberately underperforming. In litigation, where money attaches to the outcome, an unvalidated score carries little weight.
3. Error rates the vendor will not publish: Daubert asks for the known or potential rate of error. “Proprietary algorithm” is not an answer a judge accepts. Sensitivity and specificity figures from a memory clinic population also do not transfer to a 34-year-old plaintiff after a rear-end collision.
4. Silent version changes: Software updates alter the instrument. If the scoring model changed between the plaintiff’s baseline test and the post-injury retest, the comparison is not valid, and few platforms surface the version that generated a given report.
5. Test conditions nobody recorded: A tablet-based assessment done in a hallway with an interruption partway through produces a number that looks identical to one collected under controlled conditions. Paper protocols record the setting. Most digital reports do not.
6. Intended-use drift. A tool cleared as a screening aid gets cited in a report as evidence of traumatic brain injury. This is the fastest route to exclusion, and it is usually the clinician’s error rather than the manufacturer’s. Marketing that blurs the cleared indication makes it more likely.

What Device Developers Can Do

Companies that want their output to hold up in litigation can build for it:

• Publish normative data with the full demographic breakdown, including education, primary language, and age bands, not just a summary sample size.
• Report sensitivity, specificity, and confidence intervals in the populations where the device is actually used, and say plainly where the data does not extend.
• Stamp every report with the software version and scoring model, and keep prior versions retrievable.
• Log test conditions: interruptions, total duration, device type, examiner present or not.
• Document how the tool functions alongside standard performance validity testing rather than competing with it.
• Keep marketing copy inside the cleared indication. A screening aid described as a diagnostic creates liability for the clinician and exclusion risk for the case.

None of this is exotic. It is the same evidence trail that established neuropsychological instruments have carried for decades.

What This Means for Clinicians and Attorneys

A digital cognitive score is one input. A neuropsychological evaluation combines record review, clinical interview, collateral information, standardized testing across cognitive domains, and validity measures, then integrates them into an opinion an expert can defend on the stand. The expert carries the opinion, the device supplies data, and any report that reverses that order invites a motion to exclude.

FDA’s list of AI-enabled devices shows how young this space still is outside imaging. An analysis of that list through December 2025 counted 1,430 authorizations: 1,094 in radiology, 65 in neurology. FDA notes the list is not a complete census, but the concentration is clear. Cognitive assessment technology is improving quickly and the evidence base courts require is still catching up. Developers who close that gap on purpose will hold the part of the market where the stakes are highest.

FAQs

Does FDA clearance help at all in court?

It supports arguments about testing, standards controlling operation, and general acceptance. It is one input to a Daubert analysis, not a substitute for one.

Can a computerized test diagnose a traumatic brain injury?

No. Devices classified under 21 CFR 882.1470 are assessment aids. The regulation states they are not intended as stand-alone or adjunctive diagnostic devices and do not identify the presence or absence of clinical diagnoses.

Why do courts care about normative samples?

A test score means nothing without a reference group. If the reference group does not resemble the examinee in age, education, and language, the resulting percentile can misstate impairment in either direction.

What is performance validity testing?

Measures designed to detect whether an examinee gave full effort. It is standard practice in forensic evaluations because case outcomes create incentives clinical settings do not.

Should a device developer talk to a forensic neuropsychologist before launch?

Yes; the evidence gaps that surface under cross-examination are usually visible during study design, when they are still inexpensive to close.