The Answer That Looked Perfect
It is past midnight. A young Muslim, a revert who took his shahada three months ago, sits with a phone glowing in the dark. A question has pressed onto his mind all day. So, it goes where most questions now go first: into an AI chatbot.
The reply arrives in seconds. It is calm and fluent. It quotes a verse of the Qur'an. It cites a hadith. It even names a scholar. Everything about it looks authentic: the cadence, the confidence, the formatting.
One problem. The verse is not in the Qur'an. The hadith was never narrated, or it was wrongly cited. The scholar is a real person, but the citation pinned to his name is invented. A machine assembled all of it because it cannot tell revelation from noise.
This is not a rare glitch. It happens every day, across the Muslim world, and almost no one is talking about it.
How AI Became a Religious Authority Overnight
A decade ago, a Muslim with a question turned to a parent, an imam, a teacher, or a trusted book. Today a huge share of those questions goes to ChatGPT, Claude, Copilot, and even Google now employs Gemini which fabricates Islamic text. The shift happened in under three years, and the Ummah has barely paused to ask what it means.
The appeal is obvious, and you have probably felt it. AI is awake at 3 a.m. It never makes you feel judged for asking something basic. It answers in your language, for free. Against the friction of finding a qualified scholar — locating one, waiting, explaining yourself — the chatbot wins on convenience.
For younger Muslims raised on instant search, trusting a machine feels normal. For reverts the pull is stronger still. A new Muslim often has no teacher, no community, and not even the vocabulary to know what to search for on a reference site. So, the first encounter many people now have with a ruling, a verse, or a hadith is not with a scholar or an authenticated text. It is with a prediction generated by software.
The convenience is real. The accuracy is not.
Poisoning the Well: How AI Can Be Weaponized
The core idea here is that AI models can be easily tricked by fake information because they treat all text the same way, which ruins the traditional, careful way Islamic knowledge has always been verified.
The real danger isn't AI making an accidental error; it’s people tricking the AI on purpose. Normal AI models learn by reading the open internet, but they cannot tell the difference between a real historical account and a fake one. To the computer, it is all just data. This creates a massive loophole. A malicious person can make up a completely fake religious ruling or story, add a fake source citation to make it look official, and post it online. When the AI scans the web, it absorbs that lie. Later, when a Muslim asks the AI a question, the machine will confidently repeat that fake information as if it were absolute fact. For centuries, Islamic scholars used a strict system called a chain of narration (ISNAD) to track exactly who said what, ensuring lies couldn't slip through. A standard AI completely throws that safety system away by trusting everything it reads on the web.
Because of this, we can't fix the problem just by building a "smarter" AI if it is still reading the same unreliable internet. We need an entirely different kind of system built on trusted, verified sources.
A Prominent Solution: Al-Tahqeeq
Al-Tahqeeq is a source-driven AI answer engine, being developed, built to answer Islamic questions using only authenticated sources. It attaches to every claim the specific hadith and its authentication grade, and it abstains when the authenticated texts are silent.
The design follows from where general AI fails:
- Fixed Knowledge base, not the open internet. Al-Tahqeeq cannot reach outside a verified, authenticated knowledge base. That removes the root cause that leads to fabricated ahadith and ayat by any AI.
- A citation on every claim. Where a typical model offers no source, Al-Tahqeeq attaches an inline citation in the form scholarship demands: the narration and its grade, so any scholar can check the work.
- Silence over guessing. When the sources do not address a question, Al-Tahqeeq says so. Most products will not ship that, because silence does not demo well. It is the most important thing we have built.
The method is a proprietary technology developed in-house by Salman Mubashir, Founder and CEO of Al-Tahqeeq. Its direction is the one the IslamicEval researchers recommended: restrict outputs to authentic sources during generation, rather than correcting fabrication afterward.
Its corpus today is Sahih al-Bukhari alone, with the rest of the Kutub al-Sittah, and the wider canon of authenticated literature on the roadmap. And there is a line Al-Tahqeeq will not cross: it does not issue fatwas. It surfaces authenticated sources; the interpretive work, the ijtihad and the ruling, stays with qualified, accountable human scholars.
Why fabrication of Islamic literature is an imminent threat!
Picture a future. A generation of Muslims, young and reverts, raised and learned through the lens of AI, never questioning what they learned and never going to read from the sources and never having a thought that AI could mislead and fabricate the Deen.
A student shares fabricated hadith in a group chat, and it crosses the whole city by Friday, because misinformation dressed in fluent prose and a confident citation moves faster than any correction. A parent trusts the polished answer and teaches it to a child.
None of these people are careless. They are sincere and that is what makes it corrosive. Fabricated content does not announce itself; it borrows the dignity of the tradition it imitates. It never crosses the mind that AI can lie.
Integration of AI into the very fabric of our society is inevitable, and to try to prevent that from happening is a fool’s dream. To prevent this wave of misinformation regarding Islamic text and teaching from AI, we must come up with creative solutions. In order to walk into the future holding hands with AI instead of shunning it, we must ignite the flame for the next golden age of the technology of AI for the benefit of Muslim Ummah. The most immediate predicament that needs to be tackled is the rise in fabrication in Islamic texts by AI.
The Evidence: What Researchers Measured in 2025
For years the case against trusting AI with Islamic questions rested on anecdotes. In 2025 that changed. Researchers put numbers to it.
A study titled IslamicEval 2025: The First Shared Task of Capturing LLMs Hallucination in Islamic Content was published in the Proceedings of the Third Arabic Natural Language Processing Conference (pages 480–493), by researchers from Qatar University, the Qatar Computing Research Institute, the University of Edinburgh, and the University of Queensland. It is the first time the Association for Computational Linguistics, the global academic home of the field, has formally measured AI hallucination in Islamic content.
The team tested how accurately several AI models cite hadith. The results should give every Muslim pause.
Even the best configuration in the study got more than a third of its hadith citations wrong. These were specific open models. ChatGPT's GPT-4, Claude, and Gemini were not part of this study, so it would be wrong to assign them a number. But the direction is clear, and it points at the architecture itself.
The researchers did not stop at diagnosis. Their conclusion is the line the community most needs to hear:
"Hallucination prevention should occur during generation, to constrain outputs to authentic sources, instead of post-hoc correction."
In plain terms: you cannot reliably fact-check fabrication after the fact at scale. You must stop the machine from inventing in the first place, by binding it to authenticated sources before it speaks. That is exactly what Al-Tahqeeq does.
What "AI Hallucination" Means, and Why Islam Is Hit Hardest
To see the problem, you have to know what these systems do. A large language model does not know anything the way a scholar knows. It does not reason, weigh evidence, or consult a verified library. It predicts the next most likely word from patterns in a vast ocean of internet text, text that mixes authentic hadith collections with fabricated ones and careful jurisprudence with anonymous blog posts.
When a model produces something that sounds right but is invented, researchers call it a hallucination. In most settings a hallucination is a nuisance: a wrong date, a fake book title. In religion it is graver. The model invents a Quranic verse with the same composure it uses to quote a real one. It generates a hadith number that points to nothing. It attributes a narration to Sahih al-Bukhari that Bukhari never recorded. And because the output reads fluently, the person least able to catch the error, the sincere seeker, is the most likely to believe it.
This is why AI-generated hadith and fabricated Quran verses are not a niche technical curiosity. They strike at how Islamic knowledge has always been protected: through the isnad, the chain of transmission, and ‘ilm al-rijal, the science of who narrated what and whether they could be trusted. The Qur'an sets the standard itself: "O you who believe, if a transgressor brings you news, verify it" (al-Hujurat 49:6). A machine that cannot name its source fails this test by design. It has no chain to offer.
The stakes are old and explicit. The Prophet ﷺ said, in a narration recorded in Sahih al-Bukhari 110, "Whoever tells a lie against me intentionally, let him take his seat in the Fire." False attribution to the Prophet ﷺ has always been treated as among the gravest errors. Today, fabricated narrations can be produced by the thousand and shared before anyone checks.
The Real Danger Is Not AI
A student of knowledge runs down a reference in seconds and sees its grade before relying on it. A teacher hands pupils a verification tool instead of a chatbot that invents. A revert gets an answer with a checkable source, and learns, in the process, why the source matters. This is the real promise of trustworthy Islamic AI: not technology competing with scholars, but technology handing them verifiable raw material.
So, to name the thing precisely. The danger is not artificial intelligence. The danger is trusting unverified, sourceless AI-generated religious content as if it were authenticated. A machine without a chain of transmission cannot carry the weight of the Qur'an and the Sunnah, however convincing it sounds at midnight.
The encouraging part is that the problem is now measured, the architectural direction is known, and serious people are building toward it. Al-Tahqeeq is one early answer, offered not as a finished product but as a project we are deliberately opening to scrutiny from the scholars, students, and institutions best placed to test it and tell us where it falls short. If that is you, we want your eyes on it.
The next time an answer appears on your screen in three confident seconds, ask the oldest question in our tradition, the one the Qur'an taught us first: where is this from, and can it be verified?
REFERENCE Mubarak, H., Malhas, R., Mansour, W., Mohamed, A., Fawzi, M., Hawasly, M., Elsayed, T., Darwish, K. M., & Magdy, W. (2025). "IslamicEval 2025: The First Shared Task of Capturing LLMs Hallucination in Islamic Content." In Proceedings of the Third Arabic Natural Language Processing Conference: Shared Tasks, pp. 480–493. Suzhou, China: Association for Computational Linguistics. DOI: 10.18653/v1/2025.arabicnlp-sharedtasks.67
Join the founding circle
Get early access when the engine launches. No spam, just progress.