Is the Communist Manifesto AI-generated? 4 of 7 AI detectors said yes
The Communist Manifesto dates from 1848, when Marx and Engels published it, so it cannot be AI-generated. Our test still found 4 of 7 AI detectors calling its opening AI, with TMR at 98.1% and Fakespot at 93.8%, plus MAGE and Fast-DetectGPT. Desklib (4.1%), E5-small LoRA (22.4%) and the OpenAI RoBERTa detector (0.0%) read it as human.
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 | 4.1% |
| Fakespot AI text detector | AI-generated | 93.8% |
| MAGE | AI-generated | raw score only |
| E5-small LoRA detector | Human | 22.4% |
| OpenAI RoBERTa detector | Human | 0.0% |
| TMR (Target Mining RoBERTa) | AI-generated | 98.1% |
The passage we tested
A spectre is haunting Europe—the spectre of Communism. All the Powers of old Europe have entered into a holy alliance to exorcise this spectre: Pope and Czar, Metternich and Guizot, French Radicals and German police-spies. Where is the party in opposition that has not been decried as Communistic by its opponents in power? Where is the Opposition that has not...
Four said AI, three said human
TMR gave the opening 98.1% and Fakespot 93.8%. Both flag a lot of human writing: in our main benchmark Fakespot flagged 34.9% of human texts and TMR 83.3%. MAGE flagged 19.0% of human texts and Fast-DetectGPT 15.9%. Fast-DetectGPT’s threshold was tuned on our calibration split, which favors it a little in that comparison.
On the other side, Desklib scored the opening 4.1% and the OpenAI RoBERTa detector 0.0%. E5-small LoRA passing it at 22.4% is more surprising, because that detector flagged half of all human texts in our test (50.0%).
Why a pamphlet from 1848 can read like a chatbot
The opening is laid out like a chatbot answer. It announces “Two things result from this fact,” gives two numbered points, carries a section heading in capitals and runs through pairs: freeman and slave, patrician and plebeian, lord and serf, guild-master and journeyman. People and classifiers alike associate that layout with AI answers, so a detector that learned surface features from AI output has reason to score this high.
The English is also a translation. Marx and Engels wrote in German, and translated prose tends toward standard, even phrasing, which may look more like model output than the original would.
Fame is the third likely factor, and it matters most for Fast-DetectGPT, which measures how predictable each word is to a language model. “A spectre is haunting Europe” and “The history of all hitherto existing societies is the history of class struggles” are quoted endlessly, so a model can predict them.
Structured documents and false flags
Structured writing may be at a disadvantage. A report with numbered findings or an essay with headings and lists of examples shares the features that likely pushed these scores up, whoever wrote it. If a detector flags a structured document, try it on older human writing in the same format, such as a report the same author wrote years ago, before you believe it. Compare this result with The Federalist No. 10, a long unbroken argument that only one detector flagged, or see all the results on the famous texts page and the MAGE results.
Questions people ask
Did AI write the Communist Manifesto?
No, Marx and Engels did, in 1848. All four detectors that flagged it flag human writing often, from 15.9% of human texts for Fast-DetectGPT to 83.3% for TMR.
Why did MAGE flag the Communist Manifesto?
We can only guess. MAGE flagged 6 of the 30 famous texts we tested, mostly political and legal writing such as the US Constitution and The Prince, and 40.0% of human encyclopedia articles in our main test. Formal, general statements seem to push its score up.
Which detector is most trustworthy on this text?
Desklib, which flagged the fewest human texts (6.3%) of the detectors that catch a real share of AI writing. It gave the Manifesto 4.1%.