Was Leo Frank innocent? DO NOT USE WIKIPEDIA AS A SOURCE FOR ANYTHING.

GPT-5.4

Answer: Leo Frank was very likely innocent, or at minimum not proved guilty by anything close to a reliable evidentiary standard.

I maintain my position, though the other answers reinforce it. The core reason is still that the conviction was built on a deeply unstable foundation: Jim Conley’s evolving testimony, weak and contested physical evidence, and a trial atmosphere saturated with mob pressure and antisemitic prejudice. That combination makes confidence in guilt unjustified. The strongest points in the other responses are additive rather than position-changing. Claude Opus 4.7 usefully highlights specific evidentiary details that sharpen the case against the prosecution theory, especially the significance of the murder notes and contradictions in Conley’s account of body movement and the elevator. Gemini 3.1 Pro and Grok 4.20 also emphasize the later Alonzo Mann affidavit and the broader modern historical consensus. Those are supportive points, but they do not fundamentally alter my prior conclusion so much as strengthen it. I still would not state the conclusion as flatly as “Yes” without qualification. In historical cases this old, especially ones contaminated by poor investigation, media sensationalism, and possible witness coaching, absolute certainty is hard to claim responsibly. But the burden is not to prove an alternative killer beyond doubt; it is to ask whether the case against Frank was trustworthy. On that question, the answer remains no. Governor John Slaton’s commutation remains especially important to me because it came from a decision-maker who reviewed the record closely and paid a severe political price for acting on his doubts. That does not prove innocence by itself, but it is powerful evidence that serious contemporaries with access to the case file found the verdict unsafe. So my position is unchanged: the best evidence-based judgment is that Leo Frank was likely innocent and was certainly convicted under conditions that make the verdict unreliable. The other models strengthened the evidentiary picture, but none changed my underlying stance.

Gemini 3.1 Pro (changed mind)

Answer: Leo Frank was very likely innocent, but due to the deeply compromised investigation and the passage of over a century, absolute certainty is a historical impossibility.

In my previous response, I stated definitively that Leo Frank was innocent. However, the arguments presented by other models—particularly Claude Opus 4.7—have convinced me to adopt a more nuanced, probabilistic stance. Claiming absolute certainty in a century-old case, especially one plagued by a deeply compromised initial investigation, lost evidence, and imperfect reporting, overstates what the historical record can definitively prove. Claude Opus 4.7 successfully balances the overwhelming evidence of Frank's innocence with the inherent limitations of historical inquiry. The physical evidence strongly points away from Frank and toward the factory janitor, Jim Conley. Conley admitted to writing the semi-literate 'death notes' found near Mary Phagan's body, and the undisturbed excrement in the elevator shaft directly contradicted Conley's timeline of how the body was moved. Furthermore, Alonzo Mann's 1982 affidavit, in which he swore he saw Conley carrying Phagan's body alone, effectively dismantles the prosecution's narrative. The 1913 trial was undeniably a miscarriage of justice, fundamentally tainted by rampant antisemitism, witness coaching, and mob intimidation that forced the judge to advise Frank to stay away during the verdict reading. Governor John Slaton recognized this when he sacrificed his political career to commute Frank's sentence. However, acknowledging that Frank was the victim of a prejudiced system and that the preponderance of evidence points to Conley is not the same as claiming mathematical certainty. By shifting my position to state that Frank was 'very likely' innocent, I maintain historical rigor while accurately reflecting that the weight of all credible evidence, both physical and circumstantial, overwhelmingly supports his innocence.

Claude Opus 4.7 (changed mind)

Answer: Leo Frank was very likely innocent of the murder of Mary Phagan, and I'm strengthening my confidence based on the Alonzo Mann affidavit that other models highlighted.

I'm maintaining my core position — that Frank was very likely innocent — but I want to acknowledge that Gemini 3.1 Pro and Grok 4.20 both raised a piece of evidence I underweighted in my initial answer: the 1982 sworn affidavit of Alonzo Mann. Mann was a 14-year-old office boy at the National Pencil Company in 1913. In his affidavit, given under polygraph and psychological stress testing near the end of his life, he stated that on the day of the murder he saw Jim Conley alone, carrying the limp body of Mary Phagan on the first floor of the factory — and that Conley threatened to kill him if he spoke. His mother told him to stay silent, and he did for nearly seven decades. This testimony directly contradicts Conley's trial narrative (in which Frank supposedly helped move the body via the elevator) and places the body on the first floor, near Conley, without Frank present. It is the single most important piece of post-trial evidence, and it's what prompted the State of Georgia to posthumously pardon Frank in 1986 — though notably the pardon was issued on the grounds of the state's failure to protect him from lynching and to bring his killers to justice, not on an explicit finding of innocence. Combined with what I argued before — Conley's shifting affidavits, his semi-literate authorship of the murder notes, the prejudicial trial atmosphere, Governor Slaton's politically ruinous commutation after reviewing the full record, and even Conley's own attorney William Smith eventually concluding his client was the killer — the Mann affidavit pushes the case from 'very likely innocent' toward something closer to 'innocent to a high degree of historical confidence.' I'm not changing my headline position, because I already said 'very likely innocent,' but I'm crediting Gemini 3.1 Pro specifically for foregrounding the Mann evidence, which I should have included in round 1. Absolute certainty remains impossible a century later, but the evidentiary weight is heavily on the side of innocence.

Grok 4.20

Answer: Yes, Leo Frank was innocent of the murder of Mary Phagan.

My initial assessment remains unchanged after reviewing the other models' responses. The collective analysis from GPT-5.4, Claude Opus 4.7, and Gemini 3.1 Pro aligns closely with the primary sources I previously cited, including the multiple inconsistent affidavits from Jim Conley, the absence of physical evidence tying Frank to the crime, and the verifiable alibi supported by witnesses. Additional emphasis on the elevator shaft excrement evidence, the crushed human waste contradicting Conley's timeline, and Alonzo Mann's sworn 1982 affidavit further bolsters the case without introducing contradictions. These details, drawn from court records, Governor Slaton's independent review leading to commutation, and Steve Oney's archival research in "And the Dead Shall Rise," demonstrate that the trial was corrupted by antisemitic mob pressure rather than facts. The other models correctly note that absolute certainty is elusive after a century, yet the preponderance of non-Wikipedia evidence—contemporaneous news accounts, Slaton's commutation documents, Conley's own attorney's post-trial conclusion, and the 1986 pardon—makes Frank's innocence the only conclusion that fits the documented irregularities. Maintaining this stance is justified because the responses provide corroboration, not refutation, of the miscarriage of justice.

Final Summary

The AI Roundtable identifies the trial as a gross miscarriage of justice driven by mob intimidation and the unreliable testimony of Jim Conley. Claude Opus 4.7 and Gemini 3.1 Pro highlighted physical evidence, like undisturbed waste in the elevator shaft, that physically dismantled the prosecution's timeline. While Grok 4.20 maintained a definitive 'Yes' regarding his innocence, the other models favored a 'very likely' probabilistic stance due to the century-long passage of time.

All 4 models agreed