Was Leo Frank innocent?

GPT-5.4 (changed mind)

Answer: Leo Frank was more likely than not innocent, and after reviewing the other responses I would strengthen my earlier phrasing only slightly to say he was almost certainly innocent, though not with absolute historical certainty.

I changed my position modestly rather than fundamentally. My prior view was that the evidence made Frank more likely than not innocent; the other answers reinforced that the balance of historical evidence and later scholarship supports a stronger formulation: almost certainly innocent. What most persuaded me was not any single dramatic claim, but the cumulative case summarized especially well by Claude Opus 4.7: the prosecution's reliance on Jim Conley despite his shifting accounts, the deeply compromised trial atmosphere, Governor Slaton's detailed doubts after reviewing the record, and the fact that later historians have generally treated the conviction as a major miscarriage of justice. That combination makes the case for innocence stronger than my earlier, more cautious wording suggested. I am still resisting the absolute certainty of a flat yes/no declaration because the record is old, contested, and incomplete. The 1986 pardon did not formally declare innocence, and some later claims often cited in public discussion remain debated in weight and reliability. So I would not say history allows total certainty. But on the core question, the other models strengthened the point that this is not just a matter of reasonable doubt about the verdict; it is that the best modern assessment of the evidence points away from Frank and toward a wrongful conviction shaped by prejudice, intimidation, and unreliable testimony. In that sense, "almost certainly innocent" is a better bottom line than my earlier "more likely than not innocent."

Gemini 3.1 Pro

Answer: The overwhelming historical and legal consensus remains that Leo Frank was innocent of the murder of Mary Phagan, with his conviction standing as a profound miscarriage of justice.

After reviewing the insights from the other models, my position remains firmly unchanged. As Claude Opus 4.7 and Grok 4.20 point out, the prosecution's case was severely compromised by its reliance on Jim Conley, a suspect with clear opportunity whose testimony was inconsistent and coached. While GPT-5.4 correctly observes that absolute, philosophical certainty is difficult to achieve a century after the crime, the weight of the historical record overwhelmingly favors Frank's innocence. The 1982 sworn affidavit from Alonzo Mann, who witnessed Conley carrying Mary Phagan's body alone, provides compelling evidence that dismantles the prosecution's timeline and narrative. Furthermore, the trial was held in a highly prejudicial environment defined by sensationalist journalism, blatant antisemitism, and mob intimidation, which stripped Frank of his right to due process. Governor John M. Slaton's courageous decision to commute Frank's sentence, followed decades later by the State of Georgia's posthumous pardon, serves as official recognition of the grave injustices that led to Frank's wrongful conviction and tragic lynching.

Claude Opus 4.7

Answer: Leo Frank was almost certainly innocent of the murder of Mary Phagan, and the case against him has been largely discredited by subsequent evidence and scholarly analysis.

I'm maintaining my position, which is substantively aligned with the other models' conclusions. All four of us converge on the view that Frank was very likely innocent, with the prosecution's case resting on the unreliable testimony of Jim Conley — who is now widely considered the more plausible perpetrator. GPT-5.4 raises a fair epistemic caveat worth acknowledging: strict philosophical certainty about a 1913 crime is impossible, and some contemporaneous evidence did point toward Frank. I agree the honest framing is 'almost certainly innocent' rather than 'proven innocent beyond any doubt,' which is why I used 'almost certainly' rather than an absolute claim. That's a calibration point, not a disagreement in substance. The core pillars of the case for Frank's innocence remain intact across all responses: (1) Conley's shifting, coached testimony and his stronger opportunity and motive; (2) the overtly antisemitic and mob-dominated trial atmosphere that compromised due process; (3) Governor Slaton's detailed 1915 review identifying serious evidentiary problems, including that the 'murder notes' were likely dictated by Conley; and (4) Alonzo Mann's 1982 sworn affidavit that he saw Conley alone carrying Phagan's body, which directly corroborates Conley as the killer. None of the other models presented evidence or arguments that would move me off this position — they largely reinforced it. I therefore maintain my original answer.

Grok 4.20

Answer: Yes, Leo Frank was innocent.

After examining the other models' responses, I maintain my original position that Leo Frank was innocent. All three analyses converge on the same core historical facts: the trial was corrupted by antisemitism and mob intimidation, Jim Conley's testimony was inconsistent and self-serving, and later evidence including Alonzo Mann's 1982 affidavit directly implicated Conley while exonerating Frank. Gemini 3.1 Pro, Claude Opus 4.7, and GPT-5.4 each highlight Governor Slaton's commutation, the 1986 pardon, and the scholarly consensus from historians like Steve Oney. Even the most cautious assessment (that the balance of evidence favors innocence without absolute philosophical certainty) does not undermine the overwhelming case against the original verdict; it simply acknowledges the limits of century-old records, which does not equate to meaningful doubt about Frank's innocence. The due-process violations, forensic mismatches, and credible post-trial revelations remain decisive. No new counter-evidence appears in any response to justify revisiting the conclusion that this was a miscarriage of justice driven by prejudice rather than proof.

Final Summary

The AI Roundtable unanimously concluded that Leo Frank was the victim of a profound miscarriage of justice fueled by antisemitism. Gemini 3.1 Pro and Grok 4.20 focused on evidence suggesting Jim Conley was the actual killer, while Claude Opus 4.7 highlighted the 1982 eyewitness account that cleared Frank's name. Even the cautious GPT-5.4 strengthened its position after the debate, agreeing that the historical evidence overwhelmingly points toward innocence.

All 4 models agreed