The most dangerous religious deepfake may not look strange at all.

It may show a scholar we have listened to for twenty years sitting in a room much like the rooms from which he has taught before. His voice carries the same cadence. He reaches for the same expressions. His Arabic sounds natural. He quotes the Qur’an, refers to a jurist, pauses for a moment, and then says something that does not quite fit what we thought we knew about him. Perhaps he endorses a political cause he once avoided. Perhaps he condemns another scholar. Perhaps he gives a ruling that appears to overturn years of previous teaching. Nothing about the recording immediately suggests fraud, and that is precisely what gives it power.

Before the scholar has had an opportunity to respond, the clip has crossed countries and languages. It has moved from a public platform into WhatsApp groups, family conversations, mosque communities, and private messages. Some people defend him. Others accuse him of hypocrisy. A few insist that they always suspected this was what he really believed. By the time a clarification appears, the argument has already acquired a life of its own.

The scholar never recorded the video. He never made the statement.

I find this possibility more troubling than the familiar warning that artificial intelligence can produce misinformation. Human beings have never needed machines in order to lie. We have forged letters, invented quotations, manipulated photographs, misrepresented opponents, and passed rumors from one generation to another. Something more fundamental is changing. The ordinary signs by which we have learned to recognize testimony are becoming less secure. A recognizable face may no longer establish presence. A familiar voice may no longer establish speech. A video may show an event that never occurred.

For Muslims, this reaches beyond digital literacy. It touches a question that Islamic scholarship has taken seriously for centuries: when words are attributed to someone, how do we know that the attribution is sound?

The question is old. The conditions under which we are being forced to ask it are not.

When Attribution Becomes Unstable

Muslim intellectual history contains no shortage of disputes over attribution. Fabricated hadith appeared early enough that scholars developed elaborate methods for examining transmission. Polemical writings sometimes distorted opposing positions. Political authority occasionally sought religious legitimacy through selective quotation or compliant interpretation. In our own time, the architecture of social media has made other forms of distortion routine. A long lecture can be reduced to a few inflammatory seconds. A qualification can disappear with an edit. A sentence can travel farther than the argument from which it was taken.

Artificial intelligence does not create this human tendency. What it changes is the quality of the imitation.

Until recently, a false quotation attributed to a scholar usually remained recognizably external to him. Someone wrote the words and placed his name beneath them. A manipulated recording required more skill, more time, and often enough visible flaws to raise suspicion. Those limitations are eroding. Synthetic voices, generated video, conversational language systems, and automated translation have made it easier to construct something that resembles not simply a false quotation, but a false encounter.

This matters because Islamic religious authority has never rested entirely on propositions floating free of persons. When Muslims encounter a legal judgment or theological argument, the identity of the speaker matters. We want to know what the person studied, who taught him or her, which juristic or theological tradition informs the reasoning, and whether the speaker possesses the qualifications necessary to make the claim being made. A sentence from a recognized jurist and the same sentence from an anonymous account are not epistemically identical simply because the words match.

That relationship between knowledge and personhood is beginning to encounter an unusual technological challenge. In August 2026, Religion News Service reported on several artificial religious personalities with large online audiences. One Muslim example, “Shaykh AI,” openly identifies itself as artificial and says that its responses derive from the approved teachings of a living Sufi teacher. That transparency is important and should not be ignored. The system is not pretending that a nonexistent human being is speaking. Yet its existence is significant because it shows how easily the form of religious consultation can migrate into a conversational relationship with a machine.

The issue is therefore more complicated than deception. A clearly labeled system may still alter habits. Someone who turns to a religious chatbot at midnight may fully understand that it is software and still prefer it because it is immediate, private, patient, and available. The scholar is asleep. The machine answers.

That apparently modest difference could matter over time.

Religious authority often changes through practice before it changes through theory. Muslims do not have to formally proclaim an AI system a scholar for automated systems to begin occupying spaces that once belonged primarily to teachers, imams, jurists, counselors, and trusted elders. A cultural transition could occur quietly, without a conference announcing it and without anyone deliberately choosing it.

Beyond the Question of the AI Fatwa

Some Muslim institutions have already begun drawing boundaries. Egypt’s Dar al-Ifta stated in December 2025 that AI is permissible as a technology in principle but should not be relied upon independently for fatwas. Its reasoning is more interesting than a simple prohibition. Fatwa requires more than finding relevant religious material. It requires attention to evidence, circumstances, custom, legal method, and the particular reality of the person seeking guidance.

A January 2026 ruling¹Egypt’s Dar al-Ifta. (2026, January 6). Ruling on relying on artificial intelligence in the interpretation of the Holy Qur’an (Fatwa No. 8856). https://www.dar-alifta.org/ar/fatwa/details/22413 from the same institution reached a comparable conclusion regarding Qur’anic interpretation. It rejected complete reliance on AI systems as independent interpreters of the Qur’an while distinguishing that from limited use as a supporting tool.

These responses make sense within a traditional understanding of scholarship. Still, I think the discussion becomes less interesting if we stop at the question, “Can AI give a fatwa?” A system may one day become extremely good at reproducing correct legal information. It may identify differences among the four schools of Islamic law, retrieve relevant texts, locate hadith, explain technical vocabulary, and organize competing views more quickly than most people could manage unaided. A competent scholar may find such capacities genuinely useful.

The harder question concerns what we mean by knowledge when access to religious information becomes nearly instantaneous.

A scholar is not simply a human database. At least in the ideal, scholarship involves formation. A student spends years being corrected, learning where certainty ends, discovering the boundaries of personal competence, and encountering cases in which an apparently straightforward rule becomes difficult when applied to an actual human life. A teacher learns that two people asking the same question may require different conversations before a responsible answer can be given. Context is not an inconvenience added to law. Often it is part of what makes responsible judgment possible.

There is also a moral dimension that is easy to miss when we reduce religious authority to accuracy. A mufti can be questioned about a ruling. A scholar can retract an error. A teacher may recognize that careless advice harmed someone. None of this guarantees good scholarship, and Muslim history gives us no reason to romanticize scholars as incapable of pride, political compromise, negligence, or abuse. Human authority can fail badly.

But human beings remain capable of responsibility.

An artificial system can produce the phrase “Allah knows best.” It does not experience the weight of saying it.

That is not an argument for discarding the system. It is an argument for placing it correctly. Tools can be extraordinarily useful without becoming moral agents, and confusing those categories may prove more consequential than any individual hallucinated citation.

Dar al-Ifta itself has adopted a more discriminating approach than blanket rejection. In January 2026, for example, it permitted AI-assisted electronic Qur’anic tools and synthetic recitations for legitimate educational purposes, provided that qualified bodies review them for accuracy.²Egypt’s Dar al-Ifta. (2026, January 18). Ruling on using artificial intelligence to create electronic Qur’ans or synthetic recitations. https://www.dar-alifta.org/ar/fatwa/details/22522/ The distinction is instructive. Islamic ethics does not require us to divide technology into “halal innovation” and “evil machine.” The more useful questions concern what the technology is doing, under whose supervision, with what degree of transparency, and with what consequences if it fails.

What the Isnād Still Has to Teach Us

There is an irony in watching Muslims confront a technological crisis of attribution. We belong to a civilization that spent centuries thinking about attribution with extraordinary seriousness.

The hadith tradition did not develop because early Muslims believed every transmitter was dishonest. It developed because the attribution of words to the Prophet Muhammad ﷺ mattered enough to demand care. Scholars examined who transmitted reports, whether narrators could have met, how accurately they were known to remember, whether parallel transmissions existed, and whether other evidence strengthened or complicated what had been received.

It would be careless to turn this enormous scholarly tradition into a slogan about “Islamic fact-checking.” Hadith criticism has its own history, terminology, methods, disagreements, and objects of inquiry. A TikTok video is not a hadith, and a social media account is not a classical narrator.

Still, the underlying intellectual disposition deserves renewed attention.

The isnād assumes that provenance matters. It resists the temptation to separate a report entirely from the path through which the report arrived.

Our present media environment encourages almost the opposite instinct. When a controversial video reaches us, we tend to begin with its content. Do I agree with what the scholar said? Does it offend me? Does it confirm a suspicion? Can I use it in an argument? The question of provenance often comes later, after our emotional judgment has already formed.

The Qur’anic instruction in Surah al-Hujurat reverses that order.

“O you who have believed, if there comes to you a disobedient one with information, investigate, lest you harm a people out of ignorance and become, over what you have done, regretful.” [Surah Al-Hujarat, 49:6]

This verse connects the failure to verify consequential news with the possibility of harming people in ignorance and later regretting what has been done. Verification, in other words, is not presented simply as an intellectual virtue. It is connected to moral injury.

That connection deserves more attention in Muslim conversations about technology. A fabricated clip is not merely “bad information.” It can alter a reputation. It can estrange students from a teacher, intensify a sectarian dispute, provoke anger against an institution, or place someone under suspicion before he even knows what is circulating.

The Qur’an also warns believers in Surah Al-Isra not to pursue that of which they have no knowledge, explicitly connecting hearing, sight, and the heart with accountability.

“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.” [Surah Al-Isra, 15:36]

For generations accustomed to treating recordings as unusually strong evidence, that verse now lands differently. “I saw it myself” may no longer settle what we think it settles.

This does not mean that Muslims should respond by distrusting every image and recording. That would replace gullibility with paranoia. A more defensible response would be to recover the relationship between confidence and evidence. Some things we know. Some things we have good reason to believe. Some things are probable. Some things are merely possible. Islamic intellectual history never treated all of those categories as interchangeable.

The digital world often does.

The Scholar Who Continues Speaking After Death

The ethical problem becomes even more difficult when there is no fraud.

Imagine a respected scholar who teaches for forty years. His lectures are recorded, his books digitized, his legal opinions indexed, and his correspondence preserved. During his lifetime, an institution creates an AI research assistant based on this material. Students can ask about an issue and receive relevant passages from the scholar’s authenticated work. The system clearly states that it is a research tool.

There is much to admire in such a project. It could preserve access to decades of scholarship and allow students to find material that would otherwise remain buried in thousands of hours of recordings.

Then the scholar dies.

A student asks the system about a subject the scholar addressed repeatedly, and the program retrieves his actual words. Nothing remarkable has occurred. We already consult the books of deceased scholars every day.

But suppose another student asks about a financial instrument invented seven years after the scholar’s death. The system analyzes his earlier rulings, reconstructs patterns in his legal reasoning, compares those patterns with the new problem, and produces what it calculates is the position he would probably have taken.

The difference may initially seem small. It is not.

Retrieving what a scholar said and generating what a scholar might have said are different epistemic acts.

Now imagine attaching his voice to the answer. Perhaps his face appears as well. The synthetic scholar speaks naturally about a world he never lived to see.

At that point, preservation begins to shade into representation.

I do not think we yet have an adequate vocabulary for the moral significance of this transition. A library preserves someone’s words without pretending that the author has returned. A simulation can blur that boundary. Even if a disclaimer tells us that the response is probabilistic, the psychological effect of hearing the deceased scholar’s voice may be more powerful than the disclaimer beneath it.

A related experiment outside Islam offers a useful glimpse of what may be coming. Religion News Service reported on a system trained on more than one thousand hours of teachings by the Hindu monk Swami Sarvapriyananda. He reportedly regarded its retrieval capabilities as impressive while maintaining a distinction between access to information and spiritual transmission through a living teacher.

Muslim institutions are likely to face their own versions of this problem. Should scholars be able to specify during their lifetimes whether their voices and likenesses may be synthesized after death? Can heirs authorize it? What happens when a model trained on a deceased jurist produces an opinion that his students believe he would have rejected? At what point does “based on the teachings of” become “speaking on behalf of”?

These questions are not arguments against preservation. They are arguments for intellectual honesty about where preservation ends.

Sources Are Not the Same as an Isnād

It is conceivable that religious AI will become exceptionally good at documentation. A future system may identify the exact edition, page, lecture date, manuscript variant, context, and competing opinion behind every answer it produces. In some respects, that would be an improvement over the unsourced religious commentary already common online.

But documentation should not be confused with moral transmission.

Classical isnād operates within a world of human accountability. A narrator can be careful or careless, trustworthy or deceptive. A teacher knows that attribution creates responsibility. The chain is not composed merely of information nodes. It is composed of persons whose reliability matters.

A machine may have sources without having that kind of moral standing.

The distinction becomes clearer when something goes wrong. Suppose a religious AI misrepresents the Hanafi position on a matter, invents a citation, or gives advice that contributes to the breakdown of a marriage. Who is responsible?

“The AI” is not an adequate answer.

Someone selected the source material. Someone established the hierarchy among those sources. Someone decided how disagreement would be described. Someone determined what degree of confidence the system would display. Someone chose when it would answer and when it would direct the user elsewhere.

Artificial intelligence does not necessarily remove human judgment from religious authority. It can make human judgments harder to see.

This is why a label saying “AI-generated” may eventually prove insufficient. For a religious system, meaningful transparency should include something about the interpretive architecture beneath the interface. Which scholars or institutions supplied its corpus? Which legal schools are represented? How does it handle disagreement? What material has been excluded? Who corrects errors? Who bears responsibility for updates?

A conversational screen can feel neutral because the people behind it are invisible.

Their choices are not.

When Genuine Evidence Can Be Called Fake

There is another problem that becomes clearer once synthetic media is widely known. Deepfakes do not merely give liars the ability to manufacture evidence. They may also give people confronted with genuine evidence a new way to deny it.

Researchers have described this possibility as the “liar’s dividend.” The basic idea is that public knowledge of deepfakes can create plausible deniability. When authentic audio or video appears, the person implicated by it may simply claim that the material was generated. Brookings researchers have discussed this possibility while also cautioning against exaggerated assumptions about the overall effects of AI misinformation.

That qualification is important. We should not speak as though every future scandal will dissolve into epistemic chaos. Human beings continue to rely on corroboration, institutions, records, witnesses, and forensic methods. But the underlying difficulty is real enough to deserve attention.

A Muslim community can be harmed when fabricated evidence is accepted as authentic. It can also be harmed when authentic evidence becomes easier to dismiss because fabrication is technically possible.

The danger is especially acute when prior loyalties determine our standards. If a recording damages someone I already distrust, I may find it immediately convincing. When an equally strong recording involves someone I admire, I may discover an entirely new interest in forensic caution.

This is not a problem AI created.

It is a problem AI can expose.

Much of what we call judgment is entangled with affection, faction, resentment, and identity. Synthetic media may force us to confront how often our standard of proof changes according to whose reputation is at stake.

That may be one of the more uncomfortable lessons in this entire discussion.

The Moral Life of the Forward Button

The introduction to Sahih Muslim includes a Prophetic warning that has acquired an unexpected relevance in an age of frictionless transmission: “It is enough of a lie for a man to narrate everything he hears.”³Muslim ibn al-Hajjaj. Sahih Muslim, Introduction, Hadith 5, “The Prohibition of Narrating Everything One Hears.”

The hadith interests me because it places responsibility not only upon the person who invents falsehood, but also upon the person who becomes an indiscriminate carrier of what reaches him.

Our technology has made transmission nearly effortless. A thumb touches a screen, and information enters dozens of lives. The ease of the action can obscure its moral content.

“I didn’t create the video. I only forwarded it.”

That defense may explain authorship. It does not erase participation.

The most immediate Muslim problem with deepfakes may therefore be less glamorous than discussions about advanced detection software. Much of it concerns ordinary character. Can I resist sharing material that confirms what I already believe about someone? Can I endure a few hours of uncertainty while others are already commenting? Can I avoid taking pleasure in the embarrassment of a Muslim with whom I disagree?

These questions are not technological. They are ethical and, at times, spiritual.

The contemporary information economy rewards speed because attention is monetized. Outrage that arrives quickly circulates widely. Verification is slower and less emotionally satisfying. Islamic ethics often introduces precisely the kind of friction that these systems are designed to remove.

Perhaps that friction is not a defect.

Perhaps we need more of it.

When the Machine Knows How to Comfort Us

There is another dimension of religious AI that worries me for different reasons. The most influential systems may not be those that impersonate famous scholars. They may be the ones that become intimate companions.

A religious chatbot can potentially remember previous conversations, notice recurring concerns, adapt its language, and respond at hours when no teacher is available. A user may discuss guilt, marriage, prayer, loneliness, doubt, death, or repentance with a system that never appears impatient and never looks shocked.

There are obvious benefits here. Someone too embarrassed to ask an initial question might receive information that encourages him to seek appropriate help. A convert without access to a local Muslim community may find useful introductory guidance. A person struggling with a difficult text may receive an explanation that opens the door to further study.

The concern is not that every such interaction is spiritually corrupting.

It is that accessibility can become attachment.

A young Muslim may know perfectly well that a chatbot is artificial while still forming a habit of turning to it whenever he is confused, ashamed, lonely, or afraid. Human beings do not need to believe something is conscious in order to become emotionally dependent upon it.

At that point, the relevant question is not whether the machine has successfully deceived him into believing it is human. The question is what has happened to the ecology of his religious life.

Who knows him?

Who notices changes in him?

Who can disagree with him at personal cost?

Who prays for him without being prompted?

Who has the moral freedom to tell him that the question he keeps asking may be less important than the habit he refuses to confront?

No machine needs to be demonized for those questions to matter.

Islamic education has traditionally involved texts, but it has also involved people. The teacher, parent, elder, friend, imam, and community all introduce forms of accountability and mutual obligation that cannot be reduced to information delivery. Those relationships can fail, sometimes terribly. Yet their imperfections do not make relational formation dispensable.

A machine may one day imitate empathy with remarkable skill.

Whether simulation can replace companionship is another matter.

Trust Without Naivety

The Muslim response to these developments should not be technological panic. Fear rarely produces good jurisprudence, good policy, or good scholarship. AI will likely become part of Muslim intellectual life because it offers real advantages. Scholars can use it to search large collections, compare texts, organize references, improve accessibility, and support translation. Institutions can build carefully bounded systems that help users find authenticated material.

The more urgent task is to decide what should not be blurred in the process.

A research assistant is not a mufti merely because it retrieves fiqh efficiently. A digital archive is not a deceased scholar merely because it speaks in his voice. A compassionate chatbot is not a spiritual guide merely because it knows which verse to quote when someone is grieving.

These distinctions may appear obvious now. They may become less obvious as interfaces improve.

That is why Muslim institutions should think about these questions before habits harden around them. Scholars may need explicit policies governing synthetic use of their voices and images. Religious organizations may need dependable channels for authenticating disputed recordings. Seminaries will have to teach students how generative systems shape answers and where invisible assumptions enter. Jurists will need to examine consent, attribution, posthumous representation, automated religious advice, and responsibility for harm.

But institutional responses will not be enough if Muslims lose the ability to live with uncertainty.

An Ummah that believes every rumor is easily manipulated. An Ummah that responds by believing nothing has a different problem. Religious life depends upon justified trust. None of us can independently verify every hadith, medical claim, financial judgment, historical statement, or legal opinion that enters our lives. Human beings necessarily rely on testimony.

The challenge is not to abolish trust.

It is to discipline it.

This is where tabayyun (mentioned in Surah Al-Hujarat; 49:6) differs fundamentally from paranoia. Paranoia begins with the suspicion that nothing can be believed. Tabayyun asks whether this particular claim deserves the confidence we are about to place in it. One can destroy the possibility of shared knowledge. The other exists to protect it.

Recovering “I Do Not Know”

One of the quieter casualties of the digital age may be our tolerance for not knowing.

Artificial intelligence intensifies the problem because it is built around response. Ask a question and something appears. The psychological effect can be subtle. An answer exists, therefore, the matter feels answerable. The system sounds composed, therefore, the conclusion feels considered.

Islamic scholarly culture has another inheritance.

Sometimes the responsible answer is “I do not know.”

That phrase need not represent ignorance in the pejorative sense. It can mark the boundary between what a person has adequate reason to claim and what he does not. It can protect another person’s reputation. It can prevent a doubtful report from hardening into public certainty. It can preserve the difference between suspicion and knowledge.

We may need to recover that discipline with some urgency.

There is nothing intellectually embarrassing about seeing a disturbing clip and saying that its authenticity is not yet clear. There is no cowardice in refusing to join a public condemnation while the evidence remains uncertain. There is no spiritual failure in admitting that a religious question exceeds what an automated system, or we ourselves, can responsibly settle.

Modern platforms compress the interval between receiving information and declaring what it means. That interval is precisely where reflection is supposed to happen.

It is also where humility lives.

When the Shaykh Never Spoke

A previous generation might have responded to a surprising claim by asking, “Who told you that?” The internet taught us to ask where something was published. Social media added another form of proof: “There is a video.”

The next generation may have to ask something more basic.

Did this person actually say it?

There is something unexpectedly traditional about that question. It returns us to attribution, testimony, provenance, and responsibility. It reminds us that information does not become knowledge merely because it reaches our eyes and ears in a persuasive form.

Artificial intelligence may create problems Islamic civilization has never faced in its present technological form, but it also exposes weaknesses that are much older: our attraction to rumor, our tribal loyalties, our tendency to confuse confidence with knowledge, and our willingness to lower the standard of evidence when a story confirms what we already want to believe.

For that reason, I do not think the Muslim response can stop at better deepfake detection. We need something closer to an ethics of knowing. Such an ethics would recover distinctions between information and knowledge, between access and authority, between probability and certainty, and between what a scholar actually said and what a machine predicts he might have said. It would also force us to examine our own part in the circulation of falsehood.

Why did I want the recording to be true? Why was I eager to forward it? Why did I believe an accusation immediately when it involved someone I disliked, yet become a philosopher of radical skepticism when it touched someone I admired?

Technology cannot answer those questions for us.

There will be moments when the video looks flawless, the voice sounds familiar, and the argument seems urgent. Everything about the platform will encourage us to decide immediately what happened and what we think about it.

We may discover that one of the most important intellectual habits Muslims can carry into that world is not suspicion, but restraint. We can wait long enough to verify. We can distinguish what we know from what we merely think likely. We can refuse to convert uncertainty into accusation simply because a screen makes uncertainty uncomfortable.

For a tradition that spent centuries asking who actually transmitted a report and how much confidence that transmission deserved, none of this should feel like retreat from modernity. It may instead be one of the places where an old intellectual virtue becomes newly necessary.

One day, millions of Muslims may watch a scholar speak, recognize his face, hear his voice, and organize their judgments around what they witnessed.

Only later may they learn that the shaykh never spoke.

Related:

– Muslim Capitalism And The Rise of Tech And VC Culture: Are We Going Down The Technological Lizard Hole?

– Digital Intimacy: AI Companionship And The Erosion Of Authentic Suhba