Kelsey's subheadline: "Francesca Gino, Lawrence Lessig, and the threat of a false AI consensus". Larry Lessig uses "AI" to launch a devastating and successful cognitive attack on your own brain...
The story of Francesca Gino is unbelievable—not in the sense that is not true, but that she could think that this way of behaving was and is a good idea demonstrates an enormous gap between me and my ability to empathize and track her thinking. Yes, we are social network beings who stand by their friends—that is what being a friend is, and friendship is a good thing—but Larry Lessig’s place in this story is also something that I find unbelievable. Against that background, we should note and focus on the role of “AI “in this as a way not to improve your rationality, but rather to improve your rationalization of beliefs you hold for reasons that are profoundly irrational and contrary to reality…
The Gino case was already a story about academic fraud.
Larry Lessig has now turned it into a story about AI-enabled self-deception.
If you only ever give the robots your side of the story, don’t be surprised when they echo your outrage back to you, supercharging rationalization rather than rationality. The evidence with respect to Francesca Gino are damning: manipulated datasets, vanished files, and an investigative record that points squarely at intentional fraud. Lawrence Lessig has chosen to champion her anyway. Harvard’s investigation into Francesca Gino produced a 1,300‑page report, retractions, and a mountain of forensic detail pointing to data manipulation. Rather than grapple with that record, Lawrence Lessig built a long-form podcast that brackets away the ugliest facts and then turned to AI to bless his version of events, building a nine‑hour podcast that omits the most incriminating of the facts. He then fed only that podcast into ChatGPT, Claude, and Gemini and celebrated their summaries as “accurate” accounts.
And he believes it finding it “pull[ing] together the arguments in the case and presented them in a much more compelling and persuasive way than I…”
(SEMI-)CROSSPOST: KELSEY PIPER: Yes, You Can Trick Ai into Exonerating Someone

Jul 08, 2026
∙ Paid
When Harvard Business School researcher Francesca Gino was fired from her tenured job over research misconduct, she promised to take the fight to court: She sued Harvard and the researchers who outed her for defamation and (in Harvard’s case) for failing to follow its process to investigate her.
That didn’t go particularly well.
Gino’s downfall has been one of the most scintillating pieces of academic drama in recent years. Gino was a star behavioral scientist at Harvard Business School whose work often focused on honesty and ethics. She often found strikingly large effect sizes for fairly minor interventions that, in many cases, did not replicate.
For example, in one study, she checked whether signing a “statement of integrity” at the top of the page or at the bottom changed the odds that participants would cheat in an exercise where they were paid based on how many puzzles they reported solving. The effect sizes claimed in this paper are staggering: “Signing at the top vs. the bottom lowered the share of people over-reporting their math puzzle performance from 79% to 37% (p = .0013), and lowered the average amount of over-reporting from 3.94 puzzles to 0.77 puzzles (p < .00001),” Data Colada wrote.
Big if true, as they say. But it wasn’t true.
The team of external researchers at Data Colada looked at the raw data in Excel and noticed that six participants had been moved to the wrong group: Three big cheaters were incorrectly moved to the “signed at bottom” group and three honest people were moved to the “signed at top” group. Not the typical p-hacking, then, but intentional data manipulation — and Gino was reportedly the only one of the paper authors involved in data collection and analysis.
Harvard, as part of its investigation, found the data file that was emailed to Gino by a research assistant and observed that it was quite different from the one Gino sent to her coauthors days later. Gino had added three participants (who seem to have been legitimate participants who really existed) and changed a bunch of cells (including the swaps that the Data Colada team found, as well as some other changes).
Let’s look at the timeline:
In fall 2021, three behavioral scientists on the Data Colada team flagged anomalies in four of Gino’s Harvard Business School papers (including the aforementioned honesty pledge paper). Harvard investigated with an outside firm and produced a nearly 1,300-page report concluding that she “committed research misconduct intentionally, knowingly, or recklessly.”
The fallout was massive. Gino lost her tenured job, four of her papers were retracted, and she has become an object of derision. But where any other scholar might slink away in embarrassment, Gino decided to double down. Her defamation claims, which the judge called “weak indeed,” were thrown out, but the litigation over the firing process is still proceeding.
Harvard filed a counterclaim against her, and in the course of litigation, Harvard’s original report that led to her firing came out. It contained some remarkably damning details, and more details have come out in the course of the litigation that are, if true, even more damning.
The single most incriminating detail is that investigators found an early data file on Gino’s computer whose numbers didn’t match what she had published. Mid-investigation, that file vanished, and a new one was in its place with the same name but different numbers that matched the published version. Whoever swapped it tampered with the file’s hidden date stamp to make it look over 10 years old.
Harvard wants to be allowed to investigate who deleted the original file, replaced it with a different file, and backdated the replacement.
I’m not a court of law, but I have a theory of who did it: probably the same person who did all of the other research fraud.
But AI told me I was right!
Gino has taken her case to the court of public opinion. By her side has been prominent legal scholar Lawrence Lessig, who clerked for Justice Antonin Scalia and has since developed a recurring habit of taking up contrarian cases that tend to backfire. For instance, he defended an MIT administrator’s decision to take money anonymously from Jeffrey Epstein and then sued The New York Times for “clickbait defamation” when it headlined a piece: “A Harvard Professor Doubles Down: If You Take Epstein’s Money, Do It in Secret.”
Over the last year, Gino and Lessig put together a long podcast making the case for her innocence. Lessig is a talented guy, and his long and meticulous case sounds quite convincing.
The problem is that it leaves out most of the actual evidence against Gino…
Brad DeLong here: The rest of what Kelsey Piper has to say is behind the paywall. But, first, Lessig:
Larry Lessig: Bested by a Machine : ‘For almost a year now, I’ve been unpacking the story of Harvard’s decision to remove the tenure of HBS professor Francesca Gino…. We took the [podcast] transcripts… and fed them into an AI and had the AI produce its own summary or artifacts…. I was blown away by its output. Far better than I, it had pulled together the arguments in the case and presented them in a much more compelling and persuasive way than I had…. The most valuable resource that any of us has is time. There are plenty of sensible souls who might wonder about the story behind this case but who don’t have the time to listen to nine episodes of a podcast or read the thousands of pages…. Respecting that reality, these materials might provide a way in…
Now, Piper: Highlights:
Lessig asked ChatGPT, Claude, and Gemini to write “summaries of the whole case” based only on his podcast, and claimed they were reliable if you wanted to learn but did not have a lot of time.
Lessig claims the summaries give “accurate understanding of the facts”.
The AI outputs faithfully mirrored Lessig’s spin: producing polished, indignant denunciations of Harvard and proclamations of Gino’s likely innocence.
The pieces of evidence Lessig does not answer simply drop out of the narrative
Because AI can generate confident, well‑written, authoritative-sounding text, a partisan can wrap their preferred story in a veneer of objectivity—“Claude, ChatGPT, and Gemini all agree with me”—even though the models were only given one side.
Lessig’s use case is a cautionary tale, not a model.
Did Kelsey ask Lessig to respond?