In the name of Allah, the inspirer of truth
A concerning trend has emerged among many students, and even scholars of dīn and Muslim representative bodies, where there is a growing reliance on generating content using artificial intelligence. Not only for churning out articles and written materials at scale, but worryingly for disseminating Islamic guidance and issuing fatwās as well. While AI tools have proven beneficial in many technical and research-based fields like medicine, engineering, or logistics, their application in the domain of fatwā-writing is fraught with serious risks, risks that have already begun to manifest in alarming ways.
I recently encountered a case that highlights this issue starkly. A follower of the Ḥanafī school, during ḥajj was informed by their group imām that according to the Ḥanafī madhhab it was mandatory wājib to stay overnight in Mina. When challenged, the group imām confidently produced what appeared to be a scholarly answer complete with Arabic quotations from Al-Durr al-Mukhtār and Al-Fatāwā al-Hindiyyah, all seemingly validating the claim that staying in Mina is mandatory wājib. However, upon closer inspection, it turned out to be an AI-generated answer. It didn’t just misquote the ruling but the Arabic citations did not even exist, they were entirely fabricated. Notwithstanding the fact that the ruling contradicted the well-known position in the Ḥanafī madhhab, which clearly states that spending nights in Mina is sunnah, not mandatory wājib.
This is just one of many examples. Over the past few months, I have heard several reliable scholars voice serious concerns about this growing issue. It is high time we take a collective and principled stance on this pandemic before it plagues and corrupts the moral fabric of scholarly integrity.
This example illustrates a broader issue: AI models can and do fabricate references. They misquote texts, paraphrase in ways that alter legal meanings, and frequently conflate rulings from different madhhabs. It is not uncommon to see a Mālikī ruling presented as a Ḥanafī one, or vice versa, with confident citations that simply do not exist. Ask an AI for the Ḥanbalī ruling on wiping over socks and you may receive the Shāfiʿī view instead, or worse, a Frankenstein hybrid that belongs to no madhhab at all.
An incident that caused a stir involved the Muslim Council of Britain (MCB), which published a khutbah generated by AI quoting a Qur’ānic āyah complete with the Arabic, translation and reference, only for it to later be exposed as entirely fictional.
Many scholars have noticed AI tools generating ḥadīth with no sanad, no source, and no basis. They often appear authentic, complete with fabricated references like “reported by Ibn Mājah” or “narrated in Musnad Aḥmad”, only to discover no such narration exists.
AI-generated fabrications are rife in many fields. Lawyers have even landed themselves in trouble for using ChatGPT and not noticing imaginary court citations. Many people have reported inaccuracies when researching history, misdating events, incorrectly naming battles and figures that never existed.
Why does this happen? Quite simply because AI systems are not muftis. They are not even researchers. They do not understand usūl, qawāʿid, or ikhtilāfāt. Their goal is not the truth, but they are designed to be helpful. When prompted with a question, they do not default to “I don’t know” or flag uncertainty as a human mufti might. Instead, they attempt to generate what appears to be a coherent and authoritative answer. In doing so, they rely on patterns in data, not verified sources. If a question implicitly demands a ruling with references, the AI will try to produce one, even if it means inventing quotations. When you then ask it for a reference, it often fails to present a reliable one. This is because it is not bound by integrity, reliability, or even reality.
So, if you ask it for a quote from Al-Fatāwā al-Tatārkhāniyyah for example, it will generate one, even if that quote does not exist. This is not intentional deception, but it is the predictable result of a tool trained to generate language, not issue sharʿī rulings. In the AI world this phenomenon is known as hallucination, a recognised flaw in large language models (LLMs). When LLMs lack sufficient context to answer your question, they simply generate context and make things up, whether it’s a ḥadīth, Qur’ānic āyah, or quote from a fiqh manual.
The developers of these tools have been open about its limitations. OpenAI, for example, conducted an internal evaluation, which was reported by The New York Times.
They revealed that the GPT-3.5 (o3) model hallucinated in 33% of a benchmark test involving public figures, nearly double the error rate of its predecessor GPT-3 (o1). The more compact GPT-4 Mini (o4-mini) performed even worse, hallucinating on 48% of similar tasks. When tested on general knowledge, hallucinations mushroomed to 51% for o3 and a staggering 79% for o4-mini.
More worryingly, the newest and most advanced reasoning models from OpenAI, Google, DeepSeek and others are generating more errors, not fewer. Whilst various theories attempt to explain the reason behind this, there is no consensus on the root cause. Amr Awadallah, CEO of the start-up company Vectara that builds AI tools for businesses, and former Google executive, stated, “Despite our best efforts, they will always hallucinate…that will never go away.”
This unreliability is troubling in any field, but it is intolerable in matters tied to guidance and revelation. In such areas, we must exercise the highest degree of caution. The collapse of trust in scholarship begins with small cracks. A fabricated quote here, a misattributed fatwā there and soon the laity begin to question the reliability of the entire scholarly class. An overreliance on AI when producing religious material can potentially lead to catastrophic consequences.
This is a call primarily to scholars and muftis: Do not outsource your amānah! If you are using AI in any capacity, you must double-check every reference it produces. Never allow its false confidence to replace the humility of scholarly rigor. Our dīn is preserved through isnād, not algorithms. Imām Muslim mentions in his famous introduction:
…Muḥammad ibn Sīrīn, relates: “Indeed, this knowledge is religion, so look carefully at whom you take your religion from.” (Muslim, Introduction: 26)
The sacred sciences were never founded on convenience and ease. We all know the countless difficulties our predecessors have endured to make sure knowledge reaches us in its purest form. So, it is our duty to make sure knowledge is transmitted with precision and accuracy. AI is not a replacement for scholarship, nor can it ever replace the role of a scholar or mufti. A mufti is not a copy-paster. Iftā’ is a sacred trust, a transmission of the legacy of our honourable Nabī (peace and blessings be upon him). The one who speaks on behalf of the Sharīʿah bears a burden far heavier than most appreciate.
AI tools have their place, perhaps in summarising articles, suggesting structure, correcting grammar, or compiling public information. But using them to generate fatwās without verifying their sources in the actual books is a betrayal of our responsibility.
Let me be clear: relying on this technology without rigorous cross-checking is not just dangerous, it is deception.
Use AI if you must, but verify every word. Because one day we will stand in front of the Reckoner and give account.
May Allah accept all our efforts and utilise us for His service, amīn.
Mufti Zubair Patel
30/06/2025 (Last updated)
Approved by: