Is the Declaration of Independence AI-generated? 4 of 7 AI detectors said yes
No. The Declaration of Independence was adopted in 1776. In our test, 4 of 7 free AI detectors still flagged its opening as AI-generated, a majority. The surprise was Desklib, which flagged only 6.3% of human texts in our main test but gave the Declaration a 91.9% AI probability. Fakespot and E5-small LoRA 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 | AI-generated | 91.9% |
| Fakespot AI text detector | Human | 9.0% |
| MAGE | AI-generated | raw score only |
| E5-small LoRA detector | Human | 45.9% |
| OpenAI RoBERTa detector | Human | 0.1% |
| TMR (Target Mining RoBERTa) | AI-generated | 94.6% |
The passage we tested
When in the Course of human events, it becomes necessary for one people to dissolve the political bands which have connected them with another, and to assume, among the Powers of the earth, the separate and equal station to which the Laws of Nature and of Nature’s God entitle them, a decent respect to the opinions of mankind requires that...
Desklib flagged it, Fakespot and E5-small LoRA did not
Desklib, MAGE, TMR and Fast-DetectGPT called the Declaration’s opening AI-generated. Fakespot, E5-small LoRA and the OpenAI RoBERTa detector passed it.
That split runs against what you would expect from their track records. Desklib had the fewest false positives of any detector that catches a real share of AI text: 6.3% of the human texts in our main test, and only 4 of our 30 famous texts. Yet it gave the Declaration a 91.9% AI probability. Fakespot and E5-small LoRA flag human writing far more often (34.9% and 50.0% of human texts), and both passed it, at 9.0% and 45.9%. Fast-DetectGPT, whose threshold we tuned on a calibration set, flagged 15.9% of human texts and 17 of the 30 famous ones. TMR flagged 28 of the 30, so its verdict adds little.
Familiar wording, balanced sentences
The most likely cause is how familiar the text is. “We hold these truths to be self-evident” is among the most quoted sentences in American history, and the full text is reproduced all over the web. A language model has probably seen it more often than almost any other political text, so the wording looks predictable to it.
The style probably plays a part too. Jefferson and the Continental Congress wrote long, balanced sentences that move step by step from a general principle to a conclusion: people have rights, governments exist to secure them, and a government that destroys them may be replaced. That orderly argument, full of abstract nouns like Liberty, Safety and Happiness, has the even, measured tone people associate with AI essays. The passage also contains lists, such as “Life, Liberty, and the pursuit of Happiness”, and doubled phrasing like “it is their right, it is their duty”.
Formal writing and a careful detector
A low false positive rate is an average. Desklib’s few false positives in our main test came mostly from news reports, and here it flagged formal political prose. Anyone writing in a formal, argued register, such as a policy paper or a legal brief, should probably expect detectors to be less reliable on their work.
Thomas Paine’s Common Sense, published the same year and arguing the same cause, drew only one flag. The Desklib results show where else it goes wrong, and the famous texts page has all 30 results.
Questions people ask
Why do AI detectors flag the Declaration of Independence?
Likely because it is one of the most copied and quoted documents in English, so language models know its wording well, and because its formal, carefully balanced sentences resemble the polished prose AI tools produce.
Which AI detectors passed the Declaration of Independence?
Fakespot (9.0% AI probability), E5-small LoRA (45.9%) and the OpenAI RoBERTa detector (0.1%). Desklib, MAGE, TMR and Fast-DetectGPT flagged it.
Is Desklib wrong about the Declaration?
Yes. It gave the Declaration a 91.9% AI probability, its second highest score among our 30 famous texts after Genesis 1. Desklib rarely flags human writing, but this shows it can.