AI, Truthfulness and Fabricated Content

1. Fluency is not knowledge

Qur’an 17:36 forbids following what one has no knowledge of and places hearing, sight and heart under accountability. A smooth paragraph, realistic photograph or familiar voice is not proof of origin or truth.

Each gold Qur’an panel uses the locked Uthmani Arabic catalogue and separately named English translations. The page keeps six categories distinct: direct Qur’anic teaching, authentic hadith, qualified fiqh application, current technical evidence, current law or regulation, and Tawheed Club’s educational guidance. Technology changes quickly; a current technical description is not revelation, and a general fiqh statement is not an automatic verdict for every tool or use.

Tawheed boundary: اللّٰه (ALLAH) alone is the Creator, the Knower of the unseen and the source from whom all created ability and permitted knowledge ultimately come. People build machines from created matter and use faculties they did not originate. An AI system is a tool: it is not divine, prophetic, a rival creator, a moral authority or a source of revelation. Fluent output is not proof of consciousness, wisdom, truth or access to the unseen.

NIST Generative AI Profile 600-1 describes confabulation as confidently presented erroneous or false content and notes that generated systems may invent reasoning or citations. This is a property of how present generative systems can operate, not evidence that a machine intended to lie.

Human deception and model error must be separated. A model can produce a false statement without moral intention; a person can still act dishonestly by publishing it as verified, concealing material AI assistance or refusing to correct known error.

For adults teaching children: demonstrate with fictional prompts and non-animate sample media. Teach children never to enter a real name, face, voice, school, address, location, medical detail or private conversation into an AI tool. Keep adult-controlled accounts and privacy settings, supervise generated content, explain that convincing output can be wrong, and provide a simple route for reporting anything frightening, sexual, hateful, manipulative or secretive.

Evidence practice: before using or sharing a digital output, write who or what produced it, the original source, the date, the intended use, the people affected, the personal data involved and the consequence if it is wrong. Check the claim against a named primary or authoritative source, inspect the surrounding context, look for independent confirmation and record uncertainty. For a religious claim, locate the complete Qur’an passage or authentic report and ask a qualified human scholar about its application. Do not cite an AI system as though it were the source of sacred knowledge.

2. Verify before people are harmed

Qur’an 49:6 commands verification when consequential news arrives so that people are not harmed through ignorance and regret. The instruction corrects impulsive forwarding and requires attention to the reliability of the route by which information reached us.

Use lateral verification. Leave the page or chat, search for the original document, confirm the issuing organisation, compare the date and version, and read the surrounding passage. Look for a second independent authoritative source when the claim is consequential.

For Qur’an, open the complete verse and named translation and inspect context. For hadith, record collection and number and check authenticity. For a scholarly ruling, name the scholar or institution, exact question, date, conditions and recognised differences. “AI says Islam teaches” is not a source.

Screenshots are weak evidence on their own: they can omit context, show an old interface or be edited. Ask for the original URL, file or recording and determine whether the claim changed after publication.

Evidence practice: before using or sharing a digital output, write who or what produced it, the original source, the date, the intended use, the people affected, the personal data involved and the consequence if it is wrong. Check the claim against a named primary or authoritative source, inspect the surrounding context, look for independent confirmation and record uncertainty. For a religious claim, locate the complete Qur’an passage or authentic report and ask a qualified human scholar about its application. Do not cite an AI system as though it were the source of sacred knowledge.

3. Speak straight and disclose synthetic content

Qur’an 33:70 commands believers to fear اللّٰه (ALLAH) and speak words that are sound and straight. A material fact about how content was produced should not be hidden where concealment would mislead the audience.

Disclose AI assistance when it affects authorship, evidence, assessment or a person’s decision. Label synthetic or materially altered audio, images and video clearly. Do not imitate a named person’s voice, face or writing style in a way that suggests their participation or endorsement without valid permission.

C2PA Content Credentials specifications provides a technical standard for cryptographically bound provenance information and change history. It can record origin and edits, but the specification itself explains that provenance is not a judgement that the content’s claims are true. Verify the source and the underlying claim.

Keep the original, edit history and source list for important public work. If an error is found, correct the public record with comparable visibility rather than silently changing a file that has already influenced others.

Evidence practice: before using or sharing a digital output, write who or what produced it, the original source, the date, the intended use, the people affected, the personal data involved and the consequence if it is wrong. Check the claim against a named primary or authoritative source, inspect the surrounding context, look for independent confirmation and record uncertainty. For a religious claim, locate the complete Qur’an passage or authentic report and ask a qualified human scholar about its application. Do not cite an AI system as though it were the source of sacred knowledge.

4. AI must not manufacture revelation or rulings

Qur’an 16:116 forbids falsely declaring “lawful” and “unlawful” based on what tongues describe, because doing so invents a lie against اللّٰه (ALLAH). An automated answer cannot take responsibility for a fabricated fatwa.

Do not ask a model to produce a new Qur’an verse, imitate Qur’anic revelation, fabricate a hadith, decide whether a person is a believer, or issue a binding ruling from incomplete facts. A generated Arabic sentence is not revelation; a citation-looking number is not an authenticated report.

International Islamic Fiqh Academy Resolution 258 (3/26): Artificial Intelligence requires honesty, documentation and transparency and protection of religious symbols. Its contemporary resolution supports responsible use but does not elevate AI into a mufti. A tool may help locate a source only if a person opens and verifies the source.

For religious research, preserve a chain of Responsible use of tools: researcher records the question, primary sources are opened, translations are named, qualified scholarship is consulted, uncertainty and disagreement are retained, and a human editor owns every published claim.

Ruling boundary: authentic reports contain serious warnings about image-making. Qualified scholars differ over how those reports apply to photography, video, animation, digital drawing and AI-generated imagery. This course records the texts and recognised applications without declaring the disagreement imaginary or pretending that an earlier scholar ruled on a technology that did not yet exist. Ask a qualified scholar who understands the exact medium, subject, purpose, audience and harm.

5. Deepfakes and viral claims can become injustice

Qur’an 24:15 describes people receiving a claim with their tongues and speaking with their mouths what they had no knowledge of, considering it light while it was grave with اللّٰه (ALLAH). Digital speed does not make an accusation small.

Qur’an 5:8 commands justice even where dislike exists. Synthetic media must not be used to create a false confession, corrupt evidence, invent scandal or portray an opponent doing what they did not do.

A deepfake may clone face, voice or movement. Warning signs can guide caution, but visual inspection alone is not a reliable guarantee. Seek the earliest available source, independent reporting, provenance where available and confirmation from the person or responsible institution.

Do not repost harmful media merely to condemn it when reposting increases reach or exposes a victim again. Preserve evidence through an appropriate confidential route, report the content to the platform or authorities where necessary and protect the affected person’s dignity.

Digital-safety boundary: never upload passwords, private messages, confidential school or work files, intimate images, health records, children’s identifiers or another person’s data to a public AI service. Never use generated output as the sole basis for medical, legal, financial, safeguarding or religious decisions. Immediate threats, grooming, blackmail, self-harm material, sexual exploitation or credible violence require a trusted adult and the appropriate local professional or emergency route.

6. Use the six-step verification card

Step one: state the exact claim. Step two: classify text, image, audio, video or mixed media. Step three: locate the original and identify the publisher. Step four: open the cited primary source and check context. Step five: seek independent confirmation and inspect provenance. Step six: decide to publish, label uncertainty, correct or withhold.

Sahih al-Bukhari 6018 teaches speaking good or remaining silent. Silence can be the faithful decision when verification is impossible and sharing could harm someone. Being first is not worth becoming a carrier of falsehood.

Record the result in the Digital Decision, Verification and Family Safety Workbook. If the claim concerns immediate danger, preserve what is necessary and use the proper safeguarding or emergency route instead of conducting a public investigation.

For adults teaching children: demonstrate with fictional prompts and non-animate sample media. Teach children never to enter a real name, face, voice, school, address, location, medical detail or private conversation into an AI tool. Keep adult-controlled accounts and privacy settings, supervise generated content, explain that convincing output can be wrong, and provide a simple route for reporting anything frightening, sexual, hateful, manipulative or secretive.

Evidence practice: before using or sharing a digital output, write who or what produced it, the original source, the date, the intended use, the people affected, the personal data involved and the consequence if it is wrong. Check the claim against a named primary or authoritative source, inspect the surrounding context, look for independent confirmation and record uncertainty. For a religious claim, locate the complete Qur’an passage or authentic report and ask a qualified human scholar about its application. Do not cite an AI system as though it were the source of sacred knowledge.