title: The Severed Chain
date:  2026-04-22
topics: [isnad, hadith sciences, attribution, provenance, generative AI, authorship, epistemology]
words: 2059 (9 min read)

The Severed Chain

A company called Malus.sh has built a straightforward product. You give it an open-source codebase — one released under GPL or AGPL, licenses that require anyone who ships the software to publish their own modifications — and its language model reads the code, reimplements the same functionality, and hands you back a version that its authors call a “clean-room” original. The term comes from a 1982 IBM legal precedent: if an engineer never saw the source, their reimplementation is not a copy, and no license obligations travel with it. Malus charges for this through Stripe. Its founders describe the project as satire aimed at forcing copyleft into the courts. Whether the courts agree or not, the product is a working demonstration of the thing it demonstrates: in an age of generative models, the chain from source to output can be cut by machine, at price points, on demand.

The story is small in itself. The pattern it fits is not. Forty-four percent of songs uploaded to streaming services on any given day are now AI-generated, by one major platform’s own count. Academic papers on authorship attribution are a subfield; two new ones crossed the wire this week. Code, prose, music, images: the output arrives without the chain that would let anyone say who is responsible for it. That is not a side effect of the technology. For a growing number of users, it is the reason to use the technology. Plausible deniability at scale.

The Islamic tradition has seen this problem before — in its first century, in its own idiom — and its response is one of the most sustained epistemic projects any civilization has produced. The hadith sciences exist because a community that cared very much what was attributed to the Prophet, peace be upon him, decided that provenance is a condition of admissibility. What did not have a chain did not enter. What had a chain was interrogated: who each narrator was, whether their memory could be trusted, whether they met the one above them in the chain, whether their contemporaries rejected any of their transmissions. A whole literature — al-jarh wa’l-ta’dil, the critique and accreditation of narrators — grew to assess the people through whom the knowledge had moved. Biographical dictionaries filled libraries. Every major hadith collection opens with a preface naming the criteria under which its contents were admitted.

None of this was decoration. The framework was built because the Quran named the failure first.

The forbidden category

In Surat al-Isra the command is direct:

وَلَا تَقْفُ مَا لَيْسَ لَكَ بِهِ عِلْمٌ ۚ إِنَّ السَّمْعَ وَالْبَصَرَ وَالْفُؤَادَ كُلُّ أُولَٰئِكَ كَانَ عَنْهُ مَسْئُولًا

“And do not pursue that of which you have no knowledge. Indeed the hearing, the sight and the heart — about all those [one] will be questioned.” (17:36)

La taqfu — literally, do not track, do not walk behind. Do not follow a claim to its consequences when you have no basis for it. The verse closes with a triad: hearing, sight, heart. Each will be asked about. The classical commentary reads this as a judgment on conjecture — al-dhann — but the grammar is wider than conjecture. What the verse forbids is repetition without basis. You received the thing through your ear; now your tongue is about to pass it on. There is a question in between.

Surat al-A’raf makes the category sharper. The Prophet, peace be upon him, is told to announce what God has forbidden — and the list, at the end, escalates:

وَأَن تُشْرِكُوا بِاللَّهِ مَا لَمْ يُنَزِّلْ بِهِ سُلْطَانًا وَأَن تَقُولُوا عَلَى اللَّهِ مَا لَا تَعْلَمُونَ

“And that you associate with Allah that for which He has not sent down authority, and that you say about Allah that which you do not know.” (7:33)

Saying about God what you do not know is placed in the same sentence as shirk. The sentence order matters: indecency, sin, aggression, shirk, and then speaking about Him without knowledge. The classical reading does not soften this. Mokhtasar reads the final clause as applying to Allah’s names, attributes, actions, and law. Unauthorized speech about the sacred is, in this verse, not an intellectual carelessness. It is a forbidden category.

The third verse closes the triangle. Al-Hujurat 49:6 is addressed not to narrators but to recipients:

يَا أَيُّهَا الَّذِينَ آمَنُوا إِن جَاءَكُمْ فَاسِقٌ بِنَبَإٍ فَتَبَيَّنُوا

“O you who have believed, if there comes to you a disobedient one with information, investigate…”

Tabayyun — the command to verify — is the minimum ethical response to an unreliable source. The community that heard this verse did not treat it as a private recommendation. It became the structural premise of an entire science: a report cannot be admitted on its content alone; the person transmitting it must be known, and if he is not known to be reliable, the report must be checked against others.

Three verses. A diagnosis of the human failure (17:36 — we pass on what we have not verified), a theological limit (7:33 — unauthorized speech about the sacred is forbidden), and an operational command (49:6 — investigate before you act). The tradition’s answer was not a principle. It was an institution.

The warning at the root

The hadith literature on transmission opens with a warning severe enough that its shadow falls across everything after it. Abdullah ibn ‘Amr ibn al-‘As narrated that the Prophet, peace be upon him, said:

بَلِّغُوا عَنِّي وَلَوْ آيَةً، وَحَدِّثُوا عَنْ بَنِي إِسْرَائِيلَ وَلَا حَرَجَ، وَمَنْ كَذَبَ عَلَيَّ مُتَعَمِّدًا فَلْيَتَبَوَّأْ مَقْعَدَهُ مِنَ النَّارِ

“Convey from me, even a verse. And narrate from the Children of Israel without constraint. And whoever lies upon me deliberately, let him take his seat in the Fire.” (Sahih al-Bukhari; in Riyad al-Salihin, Kitab al-‘Ilm, chapter on the virtue of knowledge.)

Three clauses, three registers. Transmit what I have said. Narrate from previous communities without anxiety — there is space for other traditions. And if anyone attributes to me what I did not say, the punishment is not merely scholarly embarrassment. It is hellfire. The sentence structure raises false attribution from a matter of accuracy to a matter of eschatology.

This warning is the reason the tradition of chains became what it became. A report whose chain could not be established was, in the jurists’ phrase, ghayr maqbul — not admissible. Not because it was necessarily false, but because the cost of admitting the false under the Prophet’s name was too high to bear. Provenance was treated as a condition of safety, not a pedantic luxury.

The duty of faithful transmission

The positive duty is stated in a separate hadith, narrated by Ibn Mas’ud:

نَضَّرَ اللَّهُ امْرَأً سَمِعَ مِنَّا شَيْئًا، فَبَلَّغَهُ كَمَا سَمِعَهُ، فَرُبَّ مُبَلِّغٍ أَوْعَى مِنْ سَامِعٍ

“May Allah brighten the face of one who hears something from us and conveys it as he heard it — for many a conveyor is more retentive than the hearer.” (al-Tirmidhi, hasan sahih; Kitab al-‘Ilm, Riyad al-Salihin.)

The verb is kama sami’ahu — “as he heard it.” Not paraphrased, not summarized, not improved. The tradition’s first instinct toward received knowledge was faithfulness to form. The narrator could be less intelligent than the hearer, less articulate, less able to explain the meaning; what was required of him was transmission without addition or subtraction. Interpretation had its own discipline, practiced by its own scholars, with its own constraints. Transmission was prior. You do not paraphrase a chain.

The force of this instinct is visible in the prefaces. Imam al-Nawawi opens his Riyad al-Salihin:

أَلْتَزِمُ فِيهِ أَنْ لَا أَذْكُرَ إِلَّا حَدِيثًا صَحِيحًا مِنَ الْوَاضِحَاتِ، مُضَافًا إِلَى الْكُتُبِ الصَّحِيحَةِ الْمَشْهُورَاتِ

“I commit in it to mention only an authentic hadith from the clear ones, referred to the well-known authentic books.”

He does this before the first hadith appears. The reader knows, from page one, what kind of thing he will be reading and under what constraint. In his other famous compilation, the Forty, Nawawi opens with a hadith on the virtue of memorizing forty hadiths — and then writes that “the hadith-masters have agreed that this hadith is weak, even though its paths are many.” He cites it, uses it for encouragement, and tells the reader its chain is defective. The alternative foundation he offers — the hadiths lighublighi al-shahidu minkum al-gha’ib and nadra-llahu imra’an sami’a maqalati — are each cited as sahih. The labeling is the method. The tradition tells the reader what it knows and what it does not know about what it is giving him.

Compare this to what the current economy of text is producing. A Malus-reimplemented codebase ships without an inline comment marking which functions the model had ingested. A generated essay arrives without a footer listing the sources that shaped its sentences. A song uploaded to Deezer does not confess its model lineage. The market, in its present form, is optimized to remove precisely the information the hadith sciences were built to preserve.

The last hadith in the chapter

At the end of the same chapter of Riyad al-Salihin that opens with convey from me, another hadith from Abdullah ibn ‘Amr closes the sequence:

إِنَّ اللَّهَ لَا يَقْبِضُ الْعِلْمَ انْتِزَاعًا يَنْتَزِعُهُ مِنَ النَّاسِ، وَلَكِنْ يَقْبِضُ الْعِلْمَ بِقَبْضِ الْعُلَمَاءِ حَتَّى إِذَا لَمْ يُبْقِ عَالِمًا، اتَّخَذَ النَّاسُ رُءُوسًا جُهَّالًا فَسُئِلُوا، فَأَفْتَوْا بِغَيْرِ عِلْمٍ، فَضَلُّوا وَأَضَلُّوا

“Allah does not take away knowledge by snatching it out of people; rather He takes knowledge by taking the scholars, until, when no scholar remains, the people take ignorant heads who are asked and give fatwa without knowledge — so they are misguided and they misguide.” (Sahih al-Bukhari and Sahih Muslim; narrated by Abdullah ibn ‘Amr.)

Read this against a machine that synthesizes authoritative-sounding answers without the scholar whose learning licenses them. The hadith was not a prediction of language models. It is a structural description of what happens when a community mistakes the production of answers for the preservation of knowledge. The hadith names the moment when the question is still asked, the answer is still given, the appearance of the relationship is preserved — and the substance has been hollowed out, because the learning that would have carried the authority is no longer present in the speaker.

The classical solution to this failure was never to produce more answers. It was to produce more chains. Verify whom you are asking. Know what sits behind the word in front of you. A body of knowledge is not a pile of claims; it is a sequence of people who took responsibility for passing a thing on, each in turn.

The first step

It is tempting, faced with the Malus product and the 44 percent AI upload figure and the authorship-attribution journals, to reach for attribution as a late-stage fix. Watermarks on outputs. Labels on generated images. Regulatory requirements on streaming platforms. These are useful, and none of them reach the matter. Attribution added at the end of a pipeline that was designed to erase it is an apology, not a practice.

The tradition’s method begins earlier. Provenance is a condition of speech. You may not pass on what you do not know you received from a source you can name. When you do pass something on, you label its quality. When you compile, you state your standard before the first entry. When the standard fails, you say so. You do not conceal the gap; you mark it, so the next hand knows what it is holding.

A generation raised on stripped-chain content will find this slow. It is slow. It was slow in the third Hijri century too, when men rode for months between cities to verify a single link in a single chain. The speed is the signal. What is cheap and immediate is produced by the short circuit. What is transmitted, faithfully, in its form, labeled to its quality, with its people named — that is what the tradition counts as knowledge. The machine cannot sever that chain, because the chain is not a technology. It is a discipline.

The discipline is older than the computer and will outlast the computer. What is at stake, in the meantime, is whether the communities that inherited it remember how to use it — and whether they will let the market persuade them that the chain was always the inefficiency, and not the thing itself.

~
~
~