Is Federalist No. 10 AI-generated? 1 of 7 AI detectors said yes
Only one detector thought so, and it was wrong. James Madison wrote The Federalist No. 10 in 1787, and of the seven AI detectors we tested, just Fast-DetectGPT flagged its opening as AI-generated. Desklib (1.6%), TMR (4.8%), Fakespot (10.4%), E5-small LoRA (29.1%), the OpenAI RoBERTa detector (0.4%) and MAGE passed it.
Tested
What each AI detector said
| Detector | Verdict | AI probability |
|---|---|---|
| Fast-DetectGPT (Qwen2.5-1.5B) | AI-generated | raw score only |
| Desklib AI text detector | Human | 1.6% |
| Fakespot AI text detector | Human | 10.4% |
| MAGE | Human | raw score only |
| E5-small LoRA detector | Human | 29.1% |
| OpenAI RoBERTa detector | Human | 0.4% |
| TMR (Target Mining RoBERTa) | Human | 4.8% |
The passage we tested
AMONG the numerous advantages promised by a well constructed Union, none deserves to be more accurately developed than its tendency to break and control the violence of faction. The friend of popular governments never finds himself so much alarmed for their character and fate, as when he contemplates their propensity to this dangerous vice. He will not fail, therefore, to...
Only Fast-DetectGPT flagged it
Six detectors called Madison human. Desklib gave his opening 1.6% and TMR 4.8%, which is unusual for TMR: it flagged 28 of the 30 famous texts we tested. Fakespot (10.4%), E5-small LoRA (29.1%), the OpenAI RoBERTa detector (0.4%) and MAGE all passed it too.
Fast-DetectGPT was the exception. In our main benchmark it flagged 15.9% of human texts, a lower rate than several other detectors, though its threshold was tuned on our calibration split while theirs came from the authors. That overall rate hides a weak spot, which is academic writing. It flagged 40.0% of human academic abstracts, against 8.3% of human news reports and none of the fiction.
Formal argument is where it struggles
Madison’s opening is built from abstract nouns (faction, instability, injustice, confusion, liberty) and long, balanced sentences that set one clause against another: “the friends of public and private faith, and of public and personal liberty”. He argues in general terms and almost never gives a concrete example in this opening.
Fast-DetectGPT judges text by how predictable each word is to a language model. Careful argument in a formal register likely tends toward the expected word, because the writer wants precision more than surprise. That would make it look predictable, which the method reads as AI. The fame of the essay probably adds to the effect. It is widely taught and quoted in the United States, so a language model has likely seen its opening many times.
TMR’s pass is most likely down to formatting. The opening paragraph of No. 10 is so long that our passage ends inside it, so the text has no line breaks, and TMR’s verdicts depend heavily on them, as the paragraph break guide shows.
Advice for anyone checking essays
Our results point to formal, scholarly writing as the place where a false flag from Fast-DetectGPT is most likely, and an argumentative essay in Madison’s register sits close to it. If only one of several detectors flags such a text, give the result little weight. Hamlet’s To be, or not to be got the same single flag from Fast-DetectGPT, and the Fast-DetectGPT page covers its genre results in full. Every famous text we tested is on the famous texts page.
Questions people ask
Was the Federalist No. 10 written by AI?
No. Madison published it in 1787, in a New York newspaper, under the pen name Publius. Fast-DetectGPT flagged it, but that detector also flagged 15.9% of human texts and 40.0% of human academic abstracts in our main test.
Would a student essay in this style get flagged?
It could. Human academic abstracts were the genre Fast-DetectGPT flagged most often, at 40.0%. The other six detectors passed Madison, so a single flag from Fast-DetectGPT on an argumentative essay is weak evidence.
Why did TMR pass this text?
The passage we tested is one paragraph. TMR flagged every multi-paragraph human text in our main benchmark, and this was one of only two famous texts it passed.