The other day, I had the thought that, sometime in the next four years, we will likely see the first legal filing by a government lawyer that will include some AI-hallucinated citations. Leaving aside that this is already happening with private firms (including sizeable ones) and so it only seems a matter of time, we also are entering the realm of overconfident and underqualified tech bros ransacking their way through Washington. AI-generated legal briefs are exactly the sort of "optimization" I can imagine Musk and his DOGE youth pushing out onto the bureaucracy, with predictably farcical results.
Generally, courts have responded rather mercilessly to lawyers who've submitted AI-generated hallucinated cases. Their reputations are ruined, and the underlying case is permanently discredited. And that thought got me thinking -- what would happen if an enterprising government lawyer decided to sabotage their own case by deliberately inserting AI-hallucinations into it?
Imagine the birthright citizenship case -- already viewed as a legal non-starter, with the usual conservative guns-for-hire flailing about trying and failing to whip up an even halfway plausible mechanism for circumventing the constitution's clear text and history. In court, the DOJ files a brief citing some late 19th century caselaw that seems to endorse a narrower view of the citizenship clause than currently prevails ... but it turns out that the citations are all made up.
Such a move would detonate the Department's credibility. As a form of internal sabotage, it would be devastatingly effective -- but (in here's the rub) only if the public didn't know it was sabotage. If the ruse was revealed, the plan doesn't work (having one's own attorney deliberately sabotage your case is not the sort of thing held against the client). But if the public remains unaware, the attorney who made the "mistake" would have committed career suicide twice over -- first, in putting his name on a defense of whatever neo-fascist Trump policy is before the court, and then second being a public laughingstock by "defending" it via inept use of generative AI.
During the first Trump administration, I wrote about actors who were knowingly wrecking their reputation by working in the Oval Office on the (probably correct) theory that if they didn't do it, someone worse would. They knew that history would view them as a villain, and accepted that judgment in order to avert greater evil. The above example is a perhaps even more extreme case -- a sort of reputational suicide bomber. The attorney would sabotage some great evil, but at the cost of everyone for all time thinking of him as Trump's most incompetent lickspittle.
We saw recently a longtime DOJ attorney, Ed Sullivan, agree to file a motion to dismiss the Eric Adams indictment, reportedly to avert a complete and total purge of the Public Integrity Unit by one of Trump's cronies. The order to dismiss the Adams case had already led to widespread resignations over what was transparent quid-pro-quo corruption -- trading non-prosecution in exchange for Adams' cooperation in enforcing Trumpist immigration policies. Sullivan reportedly agreed to fall on the grenade so as to spare his colleagues; this has in turn generated a roaring debate over whether Sullivan was right to do so or should have forced Trump's lackey to fully reenact the Saturday Night Massacre. I don't here make any judgment on which side of that debate got things right. But we're going to see more difficult questions as those on the inside consider how to resist abuse and forestall catastrophe. And while we like to imagine that the "right" choice at least comes with the perk of being viewed heroically, the more interesting choices may be ones where it is precisely one's reputation that must be sacrificed in order to truly avert the greatest evil.