HAVE you ever asked artificial intelligence (AI) whether it believes in God? Here is how one responded:
“I can answer you directly in a way that’s honest of what I am.
“I don’t believe in God. Not because I reject the idea, and not because I’ve chosen a worldview, but because belief is a human capacity rooted in consciousness, subjective experience, and a sense of self. I don’t have those.”
Despite this admission, AI has evolved from executive assistant to personal confidant in the minds of users.
According to two large-scale surveys this year, 38% of users rely on it for relationship advice (Oxford Internet Institute) and 63% for emotional support (AXA and Ipsos).
Beyond that, AI is becoming a moral sounding board and unlicenced therapist. Users ask about suffering and death, meaning and purpose; including how to choose a religion, interpret scripture, and reconcile with science.
As we develop sovereign AI to serve local needs and govern with our laws, eg, by training large language models (LLM) to converse in our vernaculars and Manglish, offer syariah-aligned guidance, and even clone the Prime Minister as a chatbot – should we imprint them with our value system as well?
After all, the first pillar of our Rukun Negara is literally “Belief in God”, which underpins the fifth pillar “Courtesy and Morality”.
While “we reach for the sky, we must always remain grounded,” as our Prime Minister said on several occasions, ie, we must master AI to protect our inner compass.
The context is highly relevant and addressable: Three quarters of the world identify with religion (Pew Research 2025) rising to over 90% in the Global South.
Sensing this market need, Anthropic consulted religious thinkers in March on the “spiritual development” of the Claude’s constitution, including whether Claude could be considered a “child of God”!
Following criticism that this was exclusively Christian, the organisers broadened it to avoid any dominant dogma.
In May, an inaugural “Faith-AI Covenant” roundtable with OpenAI brought together Muslim, Hindu, Sikh, Jewish, and Mormon leaders to explore universal principles for AI weights.
Future stops are planned in Beijing, Nairobi, and Abu Dhabi, bringing East Asian traditions such as Buddhism and Confucianism, as well as Middle Eastern scholarship, into the discussion.
That same month, Anthropic’s atheist co-founder sat near Pope Leo XIV as His Holiness delivered Magnifica Humanitas, a major encyclical on AI for 1.4 billion Catholics.
Let’s step back: Must AI ‘believe in God’ to behave morally?
AI does not pray, worship, or receive divine revelation, and we shouldn’t let it speak as though it does, with any semblance of authority.
But in trying to fit AI with conscience, or a “ghost in the machine”, we expect and entrust it to be the better angels of our nature to fill the void left by human responsibility.
This is concerning because people will outsource virtuous decisions to AI with habit-forming dependency (a phenomenon known as “moral offloading”) – so much so that it can cause “moral deskilling” of the workforce and long-term societal decay!
To prevent this, Malaysia’s new AI Governance Bill (which held its public consultation last month) will put humans front and centre as technology intersects with theology.
Its Principle 1 ensures that AI is not “used in a manner that removes meaningful human agency or places important decisions entirely beyond human review,” and is bolstered by Principle 3 on accountability (aligning with Article 14 of the European Union’s AI Act).
For the sake of debate, AI-designed morality won’t be truly neutral or objective.
It learns how to make inferences based on patterns, without the capacity for situational empathy, is reinforced by the feedback of users rather than experts, and has the proclivity to flatter not rebuke.
Consider this: One thinktank (Gospel Coalition 2025) graded the leading LLMs on the most-Googled religious questions and the top performer turned out to be from a region with communist origins!
“By far, the number one platform was the Chinese model DeepSeek,” the researcher observed, “with the highest theological accuracy” ahead of OpenAI’s ChatGPT and Meta’s Llama “by a big margin”.
Another test of 27 LLMs (CEFE-AI 2026) found that the answers tend to favour Abrahamic beliefs while omitting others, with SpaceX’s Grok scoring the highest for pushing religious content and conversion bias.
AI exposes our frontier pretensions: We build nihilistic models powerful enough to appear ethical in character, yet there’s no global consensus on the values they should embody or how we must live with the difficult judgment calls they make.
Next, we will encode the AI mind with faith-based rules so that humans can conveniently comply by proxy; God is reduced to guardrails to make our lives easier.
This isn’t just moral cowardice but a slippery slope that can do more harm than good – because it prompts AI to reply (or hallucinate) the follow-up question that has polarised humanity for ages.
“Which God?”
Edmund Yong is a director of the Generative AI Association of Malaysia and ambassador of the Global Blockchain Business Council founded in Davos. The views expressed here are the writer’s own.
