AI at work

AI is a useful tool--certainly way better than boolean searching. Having tried ChatGPT, Gemini, Claude, OpenCase, and Perplexity, Claude seems to be the better solution, not just on results, but cost. It's worth it to run any result through Grammarly's AI checker. However, if you don't check the cites, the quotes, and read the case . . .

It took 2 minutes to write this query in Claude Fable 5 High

"What is your answer to this question. "Whether the Army Court incorrectly held that Mil. R. Evid. 513 is absolute and a military judge must ignore exculpatory materials that he examines in camera." Army court decision uploaded and Appellant's Supplement here at https://www.armfor.uscourts.gov/briefs/2025Term/Alfaro260170Supplement.pdf.

This is what was pumped out about five minutes later.

​My answer: No as framed, but the framing hides the real question — and on the real question, the Army Court is half right.

The QP overstates the holding. The en banc majority never held Rule 513 is "absolute." It decided the case on a narrower ground: the military judge conducted an R.C.M. 701(g)(2) screening review for non-privileged Mellette material at the victim's election, not a 513(e)(3) review, so he had no mandate — and therefore no sua sponte duty — to evaluate privileged content for disclosure or abatement. The "no constitutional exception, period" position belongs to Judge Murdough's concurrence, which drew only four of nine votes. So the absolutist holding the supplement attacks isn't actually the holding; it's a concurrence one vote short. CAAF will see that immediately, and the government will lead with it.

Where ACCA is right. The disclosure half of the argument fails on its own terms. Murdough's Brady analysis is sound: material a judge reads in camera, which the prosecution never possesses, flunks the possession element — a judge is not the prosecution team, and Shorts draws that line correctly. The Confrontation Clause argument fails under the Ritchie plurality and Fensterer: confrontation is a trial right, not a pretrial discovery engine. If the question is "must the judge disclose privileged communications over the victim's objection," the answer after the 2015 NDAA is no, and Tinsley gets that right.

Where ACCA is vulnerable. The abatement half is different in kind, and the en banc majority elides the distinction. Payton-O'Brien's remedy doesn't pierce the privilege or "read the exception back in" — it leaves the privilege fully intact and instead conditions the prosecution on it. The privilege holder keeps her secret; the government just can't have both the secret and the conviction. That structure respects Marie's hierarchy without judicially rewriting Rule 513, and it's why Chief Judge Ohlson's B.M. concurrence sketched exactly this procedure: inform the victim, seek waiver, and if refused, follow Payton-O'Brien. ACCA's answer — "nothing in the Constitution requires it" — works for disclosure but is conclusory as to abatement, because the due process claim there isn't "I'm entitled to discovery"; it's "the government may not convict me while dispositive contradiction evidence sits sealed in the trial judge's hands." Ritchie itself reserved whether an absolute privilege would survive due process scrutiny (the CYS statute there had a court-order exception). ACCA has now made Rule 513 functionally absolute, which puts it on the unanswered side of Ritchie's reservation.

The grant case. Regardless of who's right, this is grant-worthy: ACCA (Tinsley/Alfaro) and NMCCA (Payton-O'Brien/B.M.) now give military judges opposite instructions on the same recurring scenario, the scenario is structurally guaranteed to recur (Murdough concedes commingling is "virtually certain" post-Mellette), and a sitting CAAF chief judge has already endorsed the NMCCA procedure in a concurrence. A trial judge at Bragg and one at Pendleton currently operate under different constitutional rules. That's CAAF's job to fix.

One genuine weakness in appellant's position the brief doesn't confront: the perverse-incentive problem. The judge's review here was an accommodation — the victim chose judicial screening over another round with a noncompliant records custodian. If electing that accommodation triggers a disclose-or-abate tripwire, victims will never consent to in camera screening again, and judges will stop offering it. The doctrine appellant wants punishes exactly the careful procedure everyone says judges should use. Payton-O'Brien has an answer (abatement burdens the government, not the victim), but the brief never makes it.

A few flags on the supplement itself, since you'd want them flagged: it refers to "M.H." and "MT" in Part I where it means PFC ST — apparent carryover from another brief, and sloppy in a filing asking CAAF to take the record seriously. Issue V's heading reads "FAILS TO STATE AND OFFENSE." Ellerbrock is cited in text but missing from the table of authorities. And per your standing request: "she let the Mil. R. Evid. 513 cat out of the bag and into the military judge's chamber... that cat had constitutional spots" is a mixed metaphor — the idiom (revealing a secret) collides with a literal cat acquiring spots and walking into chambers. It also concedes too much: "cat out of the bag" implies the secret is out, which is the waiver argument the court just rejected.

After telling Claude the issue was the sole issue granted, this was provided.

What the grant signals. CAAF frequently grants QPs as drafted without endorsing the premise, so don't over-read it — but taking this framing rather than sanitizing it means the Court is willing to argue about what ACCA actually held, not just what it should have held. That's the first battlefield. The government's brief will open with "the Army Court held no such thing": the majority decided the case on the narrow ground that the judge's R.C.M. 701(g)(2) screening review carried no mandate to assess privileged content, and "absolute" belongs to a four-judge concurrence. Appellant's answer has to come in the first five pages, and it exists: the majority's assuming-arguendo holding — even if the judge read everything, "nothing in the Constitution requires the judge to disclose the privileged information or abate the trial" absent the victim's consent — is the absolutist rule, applied. Add Murdough's "not required, and thus not legally permitted" formulation, which four judges joined and the majority nowhere disclaims, and the premise is defensible. But it must be defended, not assumed, or the government wins on vehicle grounds before anyone reaches the Constitution.

All of this could be Bravo Sierra, so I will be interested to revisit once we get an opinion from CAAF (or modifications when the brief, Answer, and any Reply are available). And . . . . 

This Court is yet again “burden[ed] [with] addressing AI hallucinations in court filings.” Fletcher, 168 F.4th at 234. 1 It has previously acknowledged that “AI is a powerful tool, that when used prudently, provides immense benefits.” Billups v. Louisville Mun. Sch. Dist., 2025 WL 3691871, at *4 (N.D. Miss. Dec. 19, 2025) (quoting Ferris v. Amazon.com Servs., LLC, 778 F. Supp. 3d 879, 881 (N.D. Miss. 2025) (internal quotation marks omitted; emphasis in original). However, while “[g]enerative technology can produce words,” it cannot attach “… sincerity, truth, or responsibility to what it writes. That remains the sacred duty of the lawyer who signs the page.” Mattox v. Prod. Innovations Rsch., LLC, 807 F. Supp. 3d 1341, 1343 (E.D. Okla. 2025).

In this case each of the lawyers were sanctioned by a fine, order to attend CLE on AI, and removed from the case. So, both the plaintiff and the defendant lost their lawyers because all the lawyers screwed up on their use of AI. A couple were also removed from some of their cases pending before the court. ​https://www.vitallaw.com/.