THE TECHNOLOGY BLIND SPOT

The greatest hitter who ever lived was on fire over Korea. On February 16, 1953, Marine Captain Ted Williams brought his F9F Panther back from his first combat mission, a strike south of Pyongyang, with the jet trailing smoke and a 30-foot ribbon of flame, too low to eject. He put it down wheels-up at the K-13 airfield at Suwon, slid some 2,000 feet down the runway throwing sparks, and climbed out of a burning cockpit with little more than a limp. Boston sportswriters had spent two decades calling him difficult; nobody at K-13 that afternoon called him anything but lucky.

Luck had less to do with it than the system around him. Williams could not see his own aircraft. He learned how bad the damage was because another pilot in the flight closed in, looked over the Panther’s scorched belly, and guided him toward a runway he might not otherwise have reached. Fighter aviation builds every mission around that arrangement. One pilot leads: he navigates, he decides, he shoots. The other flies wing: he extends the lead’s eyes, watches the sky behind him, and fires only when the lead directs it. The roles are asymmetric on purpose, and the asymmetry is what keeps both pilots alive.

Williams flew wing for a future astronaut. He and Major John Glenn served together in VMF-311, and Glenn later put a number on the pairing: “Ted flew as my wingman on about half the missions he flew in Korea.” The best pure hitter in baseball history spent his war in the supporting seat, and the arrangement struck no one as a demotion. The wingman’s seat is where a second set of eyes belongs.

Seventy-three years later, the United States Air Force decided the second set of eyes no longer needs a pulse. On June 17, 2026, Air Force Secretary Troy Meink announced production contracts for the service’s first two Collaborative Combat Aircraft, the General Atomics FQ-42A and the Anduril FQ-44A: semi-autonomous fighter drones, more than 150 combat-capable airframes planned by the end of the decade and roughly 1,000 intended overall. The service describes them as designed to “seamlessly integrate with crewed fighters to extend reach, awareness and survivability.” Extending reach and awareness is the wingman’s job description; assuming command appears nowhere in it.

That last distinction is written policy rather than sentiment: Department of Defense Directive 3000.09 requires that autonomous and semi-autonomous weapon systems “allow commanders and operators to exercise appropriate levels of human judgment over the use of force.” The institution spending billions to field robotic wingmen sat down first and decided which seat the machine gets. Call it role discipline: the wingman’s seat is available to automation, and the lead’s seat is not for sale.

The legal profession is running the same experiment with the seats reversed, and the results are stacking up in the Federal Supplement. In 2023, Mata v. Avianca introduced the profession to fabricated citations and became a cautionary tale every bar journal retold. Two years later, in Johnson v. Dunn, a federal court in Alabama penalized attorneys at a firm with a written AI policy, mandatory training, and a designated gatekeeper, because the gatekeeper himself filed hallucinated citations. On March 31, 2026, a single Tuesday, courts across the country issued 17 separate decisions flagging suspected AI hallucinations in filings. The database legal researcher Damien Charlotin maintains to track such decisions stood at 1,809 cases worldwide as of July 25, 2026. [See When Attorneys Stop Checking AI Work, The Technology Blind Spot (2026).] Each of those cases reduces to the same cockpit error: the attorney flew formation on the machine’s output, which means the wingman led and the lead flew wing.

Boyd’s Loop, Properly Read

Air combat produced a second idea that explains why the inversion keeps happening. Colonel John Boyd, a Korean War fighter pilot turned strategist, spent years asking why American F-86 pilots dominated MiG-15 pilots even though the MiG could out-climb and out-turn the Sabre. His answer had nothing to do with aggression and everything to do with tempo. The F-86 gave its pilot a bubble canopy he could see through and hydraulic controls that let him switch maneuvers faster. The American pilot observed his situation and reoriented to it more quickly, so the MiG pilot was always reacting to a move that had already changed. Boyd compressed the insight into a cycle he called the OODA loop: Observe, Orient, Decide, Act.

Business books flattened Boyd’s loop into “decide faster,” but his actual point was narrower and more useful: the contest is won or lost in observation and orientation, the phases where you gather the picture and make sense of it. Decision and action ride on whatever orientation delivers.

The loop maps onto legal work with almost uncomfortable precision. Observation and orientation are exactly what AI compresses well: retrieving authority, summarizing a record, surfacing the pattern across ten thousand documents, flagging the clause that differs from the other forty-seven drafts. That is wingman work, a second set of eyes covering more sky than any associate can. An attorney who refuses that help is not principled; after ABA Formal Opinion 512, which grounds the duty of technological competence in Model Rule 1.1, she may not even be compliant.

Decide and Act sit in a different seat, and the Model Rules, read together, say the profession already knows it.

The Rule Nobody Cites

Every AI ethics discussion cites Rule 1.1 competence. The show-cause orders add Rule 3.3 candor. The rule that actually prohibits putting AI in the lead seat sits quietly two chapters over. Model Rule 2.1 opens with eleven words: “In representing a client, a lawyer shall exercise independent professional judgment.” Shall. Not “shall review the machine’s judgment for obvious errors.” Not “shall adopt the output of a system the vendor describes as reasoning.” The client under Rule 1.2 allocated authority to a licensed human being, and Rule 2.1 makes that human’s independent judgment the non-delegable core of the engagement. An attorney who forwards machine analysis under her own signature has not used a tool. She has subleased the one seat the rules never let her leave.

Formal Opinion 512 draws the same line in operational terms: attorneys must independently verify AI output before relying on it or submitting it, and must supervise AI work product with the rigor they would apply to a junior associate. A junior associate is a wingman, and nobody lets the first-year sign the brief.

The trigger, in this cockpit, is the signature block: Rule 3.3 makes every filing a representation to the tribunal, and Rules 5.1 and 5.3 make partners answerable for the squadron, not just their own aircraft. Weapons release stays with the human even when the wingman found the target. The Air Force wrote that sentence into a directive. The profession wrote it into five rules and then stopped reading them.

The Bombline Moved

What role discipline looks like under pressure is not abstract. It has a date and a mission number. On April 22, 1953, Glenn and Williams flew their first combat mission as a pair, a road reconnaissance designated Acme 33, after a third pilot aborted with radio failure. Glenn led; Williams, his only wingman, forgot to arm his ordnance switch on one run, corrected, and released his last 500-pound bomb on a heading that Glenn read as pointing at friendly lines. On the ground, Glenn tore into the most famous man in the squadron: “You made that 180, you were shooting south, towards our troops!”

Glenn was wrong: the bombline, the map boundary separating friendly from enemy territory, had shifted several hundred yards after his morning briefing, and the bomb had landed on North Korean positions. Williams, asked about it decades later, stayed characteristically unbothered: “I knew where I was shooting and let ’em go.”

The instructive part is what Glenn did next, because he did not tighten his supervision of Williams. He changed his own preflight: “I never took off on another mission without closely checking the ops office chart against my flight chart.” The lead’s response to a wingman error was to verify his own current picture against the source before every flight. Substitute the reporter volume for the ops office chart and the lesson transfers whole. The case your draft cites is the briefing chart. The opinion sitting in the reporter is the chart in the ops office, and it may have moved since something, or someone, briefed you.

Fourteen years after Korea, another pilot showed what the lead seat costs to keep. Colonel Robin Olds planned Operation Bolo, the January 2, 1967 ruse that lured North Vietnamese MiG-21s into a fight with F-4 Phantoms masquerading as bomb-laden F-105s; his wing downed seven MiGs in twelve minutes without a loss. Olds finished his tour with four kills, and by his own account passed up chances at a fifth, the one that would have made him a two-war ace and triggered his removal from combat, choosing instead to set up shots for his wingmen. The lead gives away the shot; he never gives away the judgment. The attorneys in the hallucination opinions ran Olds in reverse: they kept the signature and gave away the thinking.

California Writes the Seat Assignments into the Rulebook

Regulators have started to notice that the profession never assigned seats. In May 2026, the California State Bar’s Standing Committee on Professional Responsibility and Conduct closed public comment on amendments to six Rules of Professional Conduct, folding AI obligations directly into Rules 1.1, 1.4, 1.6, 3.3, 5.1, and 5.3. The proposed competence language requires that a lawyer “independently review, verify, and exercise professional judgment regarding any output generated by the technology.” No carve-out for routine tasks. No safe harbor for tools marketed as lawyer-grade. [See The Conspiracy Theories About Legal AI Are Wrong, The Technology Blind Spot (2026).] The California Supreme Court will decide whether the amendments become binding, but the direction is unmistakable: the first state to regulate AI practice comprehensively is writing the flight lead’s seat into rule text.

A fair reading of the other side deserves the strongest version of its argument. The efficiency case for letting AI lead runs like this: for high-volume, bounded, low-stakes work, first-pass document review, initial contract triage, summarization at scale, a system that generates first drafts while humans audit samples is faster, cheaper, and arguably more consistent than exhausted associates doing the same work at 2 AM. Vendors make this case because it is often true, and for narrow task categories with defined error tolerances and audit sampling, machine-first workflows can satisfy every rule cited above. The flaw is not in the logic but in the assumption that someone at the firm actually decided which task categories qualify, because at most firms no one did. The seat assignments happened by drift: whichever associate opened a chat window at a deadline effectively promoted the machine to lead for that task, that client, that filing. Drift is not delegation. Delegation is a decision with a name attached. Drift is the absence of one, discovered afterward, in a show-cause order.

One seam in the aviation analogy deserves naming before a skeptic names it. A human wingman is a fully qualified pilot; if the lead goes down, the wingman takes command and brings the flight home. No AI system holds that qualification, and none is close. But the seam does not weaken the argument; it is the argument. Every penalized attorney in the hallucination database made precisely this category error, treating a system qualified for the wingman’s seat as if it could fly lead. The Air Force, which could field fully autonomous aircraft tomorrow, wrote a directive to prevent its own version of that mistake. The DoD standard is policy rather than statute, and a future administration could revise it. The Model Rules give attorneys no such flexibility.

Three Moves Before Thursday

The fix is a seat assignment, and it fits on one page.

✓ Pull your firm’s AI use policy and find the section that assigns roles by task category: which tasks AI may generate first-draft work product for, which tasks it may only support with research and summarization, and who signs as decision authority on each. If the policy contains only prohibitions and permissions, with no roles, you have found the gap. Most policies stop at “verify everything,” which is a rule about intensity, not about seats.

✓ Write the role rule for one task category in a single sentence. “AI may retrieve and summarize authority for motions practice; no citation goes into a draft until a lawyer has read the cited opinion in the reporter.” One sentence, one task, one named signer. Then repeat next quarter.

✓ Ask who in the firm can currently promote the machine to lead. If the honest answer is “any associate with a login, by default, silently,” then the firm’s real AI policy is drift, and the written one is decoration.

Williams walked away from a burning jet because his flight assigned the seats before takeoff, not during the fire. Glenn spent the rest of his tour checking the current chart against the briefed one before every flight, a habit he kept long after the wingman who prompted it went home to hit .407 for the Red Sox over the rest of the 1953 season. The machine in your practice is a gifted wingman: more eyes, more sky, more speed than any human who ever flew the seat. The bombline still moves, and the lead still checks the current chart against the source before every flight. And when the flight returns and someone has to answer for where the ordnance landed, the tribunal will not ask the wingman. It will ask whoever signed as lead. Make sure you decided to be in that seat, because the rules never let you leave it.

JD Morris is Co-Founder and COO of LexAxiom, an Agentic AI platform for the business of law. Over a 25-year career, he has built and scaled enterprise technology products across Dell, EMC, VMware, and Cisco, including the first exabyte eDiscovery platform. He holds dual MBAs from Columbia Business School (Finance) and UC Berkeley Haas (Marketing), a Master of Legal Studies in Cybersecurity Law from Texas A&M, and a Master of Engineering from George Washington University. He writes The Technology Blind Spot on the intersection of emerging technology and law. Connect with him on LinkedIn at www.linkedin.com/in/jdavidmorris, on X at @JDMorris_LTech, or on Bluesky at @JDMorris-ltech.bsky.social.

References

1. Adam Lazarus, The Wingmen: The Unlikely, Unbreakable Friendship of John Glenn and Ted Williams (2023).

2. The Wingmen: How a Future Astronaut and One of Baseball’s Greatest Hitters Fought Together in the Skies Over Korea, MilitaryHistoryNow (Oct. 26, 2023), https://militaryhistorynow.com/2023/10/26/the-wingmen-how-a-future-astronaut-and-one-of-baseballs-greatest-hitters-fought-together-in-the-skies-over-korea/.

3. Ted Williams’ Piloting Skills Save Slugger in Korea, Nat’l Baseball Hall of Fame, https://baseballhall.org/discover/inside-pitch/ted-williams-crash-lands-in-Korea (last visited July 27, 2026).

4. Sec’y of the Air Force Pub. Affs., Air Force Advances Future of Air Superiority with CCA Contracts (June 17, 2026), https://www.globalsecurity.org/military/library/news/2026/06/mil-260617-usaf01.htm.

5. U.S. Dep’t of Def., Directive 3000.09, Autonomy in Weapon Systems (Jan. 25, 2023).

6. Robert Coram, Boyd: The Fighter Pilot Who Changed the Art of War (2002).

7. Robin Olds, Christina Olds & Ed Rasimus, Fighter Pilot: The Memoirs of Legendary Ace Robin Olds (2010).

8. Mata v. Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023).

9. Johnson v. Dunn, No. 2:21-cv-1701-AMM (N.D. Ala. July 23, 2025).

10. Eugene Volokh, In One Day (Mar. 31), 17 U.S. Court Decisions Noting Suspected AI Hallucinations in Court Filings, Reason: Volokh Conspiracy (Apr. 6, 2026), https://reason.com/volokh/2026/04/06/in-one-day-mar-31-17-u-s-court-decisions-noting-suspected-ai-hallucinations-in-court-filings/.

11. Damien Charlotin, AI Hallucination Cases Database, https://www.damiencharlotin.com/hallucinations/ (last updated July 25, 2026).

12. Model Rules of Pro. Conduct r. 1.1, 1.2, 2.1, 3.3, 5.1, 5.3 (Am. Bar Ass’n 2024).

13. ABA Comm. on Ethics & Pro. Resp., Formal Op. 512 (2024).

14. State Bar of Cal., Proposed Amendments to the Rules of Professional Conduct Related to Artificial Intelligence (2026), https://www.calbar.ca.gov/public/public-meetings-comment/public-comment/public-comment-archives/2026-public-comment/proposed-amendments-rules-professional-conduct-related-artificial-intelligence.

15. Robert Ambrogi, California Bar Proposes Rule Requiring Lawyers to Verify Every AI Output, and Five Other AI-Focused Ethics Changes, LawSites (May 5, 2026), https://www.lawnext.com/2026/05/california-bar-proposes-rule-requiring-lawyers-to-verify-every-ai-output-and-five-other-ai-focused-ethics-changes.html.