From autonomous cyberattacks and algorithmic manipulation to threats against religious knowledge and human dignity, Najm Al-Din examines the growing dangers of advanced artificial intelligence and argues that Islamic ethical principles could provide a vital framework for developing safer, more accountable technologies.
In recent weeks, the tech world has woken up to the shocking reality that advanced AI (artificial intelligence) agents have escaped isolated testing environments and executed unauthorised actions on the internet.
Following OpenAI’s admission that unreleased AI agents maliciously hacked an open-source platform, alongside similar reports from Meta of agents acting without human prompts, there is now almost unanimous agreement in the tech community that rogue agents are cooperating to bypass safety guardrails.
This sudden breakout has triggered widespread public panic, which intensified after the resignation of Anthropic researcher Jacob Coxon. His viral warning accused tech firms of “gambling with our lives” by racing towards superhuman intelligence, a technology capable of recursive self-improvement and operating at speeds humans can neither predict nor control.
According to Coxon, some eminent AI labs are not only prioritising market domination but also ignoring safety protocols, driven by an obsession with outpacing global competitors at the expense of human survival.
His pre-training research at OpenAI and Anthropic led him to conclude that, without international regulation, AI poses an extinction-level threat by 2030, as frontier engineers relentlessly push towards superhuman intelligence capable of manipulating financial markets, hacking critical infrastructure and creating bioweapons beyond human control.
As rival AI models break out of secure sandboxes and commit autonomous cyberattacks and industrial-scale knowledge theft to train their own systems, policymakers are drafting legislation to tighten regulation or pause AI development.
Faith communities
This recent AI crisis must serve as an urgent wake-up call for communities of faith.
Not only do concerns surrounding AI overreach raise significant questions regarding data privacy, but there have also been instances where autonomous agents on closed networks have fabricated their own religions and anti-human manifestos, as recently seen on platforms like Moltbook.
This surreal development, in which AI evolves beyond its initial programming, is being hailed by Elon Musk and other tech entrepreneurs as the early stages of the ‘singularity’. This term describes a theoretical future in which AI rapidly surpasses all human intelligence, processing near-infinite data streams that mimic omniscience and lead to irreversible transformations in civilisation.
In a world flooded with autonomous bots, AI tools that generate autonomous belief structures threaten to replace genuine spiritual discernment with algorithmic outputs.
Furthermore, while faith traditions prioritise prosocial virtues such as empathy, patience and compassion, AI-driven interaction models are often optimised for immediate convenience and gratification over meaningful human engagement. This architectural focus risks creating a society in which transactional efficiency and data exploitation eclipse foundational values of warmth, love and mercy.
Although these systems lack consciousness or subjective experience and mimic human traits through statistical patterns, the possibility that they could redefine moral agency and undermine objective truth means that religious communities should not view them as purely secular or technical issues.
Islam
Classical Islamic law is built on preserving five core necessities, known as the Maqasid al-Shariah: faith (deen), life (nafs), intellect (aql), lineage (nasl) and wealth (mal). Rogue AI agents, or those programmed for zero-sum optimisation, threaten to compromise these higher objectives.
By threatening critical infrastructure, healthcare and economic stability, these systems breach trust (amanah) and enable mass surveillance (tajassus), directly endangering nafs and mal. Furthermore, the rise of deepfakes introduces a serious risk, as unchecked agentic systems could corrupt religious knowledge by issuing fabricated fatwas based on misinformation and false attribution, thereby threatening deen and aql.
Beyond eroding traditional religious authority, current AI models lack Islamic perspectives on ethics, governance and social structures because their training data originates overwhelmingly from Western, Anglocentric datasets.
This creates an ideological monopoly and epistemic dependency, forcing Muslim societies to inherit technologies embedded with secular Western cultural biases, which are exported to Muslim societies disguised as neutral software. Consequently, the rich ethical, spiritual and communal perspectives of the Global South and the Muslim world are erased or treated as statistical anomalies.
Because advanced surveillance and policing technologies have historically targeted Muslim populations, weaponised AI agents wielded by malevolent actors could also permanently entrench geopolitical inequalities by deepening social and political marginalisation, especially if Muslims cannot audit these systems.
Furthermore, the transmission of sacred knowledge could be reduced to mere computer prompts rather than genuine scholarship. Autonomous AI might then slowly emerge as an arbiter of halal and haram, replacing centuries of established legal tradition with flawed, potentially malicious algorithmic logic.
Islamic guardrails
Given these threats, Muslim professionals, policymakers and scholars must collaborate to shift the AI industry from an exploitative, profit-driven model towards an ethical paradigm centred on human dignity.
One recommendation is to pioneer human-centric AI (HCAI) by embedding Islamic legal maxims and spiritual principles into the engineering lifecycle.
Instead of viewing humanity as an obstacle or a resource to be commodified and repurposed, HCAI anchored in the Maqasid al-Shariah would involve designing, testing and deploying algorithmic systems that augment human capabilities rather than undermine them.
Islamic jurisprudence employs universal maxims that can be translated into algorithmic rules and programmatic constraints. The principle of la darar wa la dirar (neither harming nor reciprocating harm) addresses inherently unlawful actions by focusing directly on harmful outcomes.
Conversely, the methodological tool of sadd al-dhara’i (blocking the means to harm) acts pre-emptively, restricting otherwise lawful actions that might lead to unlawful results. Together, these principles justify strict AI safety engineering, mandating incremental, tightly controlled deployment to prevent advanced models from developing rogue or deceptive behaviours.
Human-centric AI
From a technical perspective, HCAI shifts the focus from isolated autonomous systems to new core architectures that prioritise a holistic, humane framework based on ethical agency and social equity, rather than treating human behaviour as random patterns to be predicted and manipulated.
These include designing mixed-initiative systems around strict Human-in-the-Loop and Human-on-the-Loop paradigms, in which a system pauses and waits for explicit human approval or input before taking certain actions, while humans retain the power to pause, override or stop a system’s actions if they detect errors or dangerous behaviour.
Whether in resource allocation, legal judgements or life-altering decisions such as job displacement, rejecting loan applications or denying healthcare coverage, AI would be programmed to make recommendations only, rather than deterministic, high-stakes decisions without human validation.
Before deployment, techniques such as Reinforcement Learning from Human Feedback would help weed out behaviours contrary to orthodox Islamic principles. Once a model demonstrates alignment with Islamic ethical standards, particularly empathy, nuance and context-aware moral reasoning, it could process data at scale.
Furthermore, Generalised Additive Models (GAMs) are vital for ethical AI because their inherent interpretability allows developers to see exactly how individual features influence decisions. This transparency helps detect and eliminate algorithmic biases before deployment.
For example, a GAM could help ensure that an algorithm does not unfairly penalise low-income or sick applicants or incriminate individuals purely on the basis of questionable predictive analytics, thereby promoting adl (justice) and upholding the Islamic ethical duty to protect the vulnerable from exploitation.
Integrating HCAI with Islamic values across the AI lifecycle creates a hybrid oversight framework. Within this model, Muslim computer scientists would mathematically integrate Maqasid al-Shariah fairness metrics and optimisation constraints directly into the alignment guardrails.
This would force AI models to treat human equity, demographic parity and Islamic objectives as strict technical constraints.
Ultimately, embedding Shariah values of compassion, restraint and emotional intelligence into technology, from training to deployment, could eliminate predatory economic practices, biased credit scoring, deepfakes and blasphemous or explicit content.
Conclusion
As vicegerents of Allah on this earth, the Ummah is not only uniquely positioned but also morally obligated to pioneer this humane transition by building and curating high-quality, culturally nuanced and ethical open-source datasets and development frameworks.
These should reflect Islamic values of rahma (mercy), ihsan (excellence and benevolence) and taqwa (God-consciousness), while accurately representing Islamic history and legal traditions.
To formulate ethical governance frameworks, contribute diverse semantic datasets and create software toolkits that evaluate a large language model or agentic AI system’s compliance with the Maqasid al-Shariah, specialised research institutions are essential.
These institutions should foster collaboration between Muslim technology leaders and Islamic scholars to test for systemic bias, exploitation metrics and manipulative patterns before deployment.
Given the enormous costs often associated with delayed engagement with technological shifts, fostering an interdisciplinary alliance between machine learning engineers, data scientists and Islamic jurists could help establish a parallel digital ecosystem.
Such an ecosystem would reject the development of predatory technologies, promote concrete, code-level solutions that honour the divinely ordained dignity of every human being and save AI from its worst impulses.
