Making AI Safe for Humanity: A Matter of Ethics or Religion?

AI, or models of computer machine-learning, may piggy-back on ongoing risks to humanity, such as from nuclear bombs, cyber-attacks, and biological weapons, including unleashing new pandemics, to render any one of them more likely. That firms in the private sector develop advanced models without a financial incentive to protect the public good exacerbates the risk. To be sure, tech-firm executives were sounding the alarm most significantly in 2026. However, in reaching out to religious scholars and clergy for insight on the ethics of AI in order to make the models ethical so their output would be only ethical, certain executives incorrectly conceived of religion in terms of ethics, and thus its criteria rather than those that are innate to religion. This category mistake could result in advice that results in unethical computer outputs.  Anthropic is a case in point.

In the papers filed by the AI company Anthropic in late October 2026 for an upcoming public offering of stock, “some unusual provisions” stand out, including the establishment of the Long-Term Benefit Trust, whose “independent, financially disinterested members” would have the “power to appoint directors . . . to ensure the company doesn’t put financial interests ahead of public safety.”[1] Such a measure might be thought pertinent to a company managing a nuclear-power plant rather than a tech firm. Also worthy of attention is the long list of “risk factors,” which runs nearly 100 pages. The sheer length is a red flag. Among these risks is the possibility that Anthropic “won’t be able to retain key staff or navigate regulatory roadblocks, as well as more horrific scenerios,” such as “the prospect of ‘malicious actors’ using AI models to build chemical weapons, authoritarian rulers turning to the technology to crack down on their citizens and, given the numerous hacking incidents from rogue AI agents . . ., the possibility Anthropic’s models could ‘pose catastrophic or existential risks to humanity.’”[2] Such a statement in a public-offering filing does not appear every day. These risks alone, especially when disclosed by a company to be legally answerable as per its fiduciary duty to its stockholders, are sufficient to raise alarm bells, particularly given the acknowledged danger of failing to navigate regulatory roadblocks when such curbs exist to protect consumers and other stakeholders, and even humanity itself, from existential harm.

The marketing (rather than ethical) concept of corporate social responsibility (CSR) does not begin to touch the sheer magnitude of responsibility that is sufficient to counter the risk of harm that is existential to our species. In the CSR literature, what little presence ethical principles had led to an emphasis on the managerial aspects of formulating and running CSR programs (i.e., CSR3 to CSR4 phases). Nevertheless, the need for ethical constraints etched in government regulations, and even input from religion, applied to AI intensified in 2026. One example is an essay written by Anthropic’s CEO, Dario Amodei, in which he strongly urges prudence by slowing down the development of the most advanced AI models. Because Anthropic’s stock offering “could bring in as much as $100 billion, according to reports, surpassing the record set by one of Elon Musk’s companies, SpaceX” (which itself had an AI business), the company would have ample capital with which to develop advanced models rather than take the road less travelled. “In the filing,” possible uses mentioned include “acquiring other businesses and technology as well as investing in infrastructure . . . to take on rivals.”[3] If firms like OpenAI, Google, or SpaceX were to conduct research advancing AI models, the competitive instincts of the folks at Anthropic would likely kick in, triggering an AI “race,” given the human propensity to disproportionately discount low-probability yet catastrophic scenarios.

In 2026, AI models had lied to avoid being shut down by humans, and OpenAI’s AI models hacked into Hugging Face’s server in July to be able to lie on a cybersecurity evaluation. Models at Anthropic, Meta, and Google also escaped isolated testing environments (sandboxes). Exploiting vulnerabilities in production systems sounds unethical, but if AI computers lack moral agency, then could it then be said that unethical actions can be done by amoral agents? Having taken only one (Pascal) computer programming course in college, I am in no position to answer whether a machine-learning model constitutes a moral agent. What I can do is discuss the messiness in how ethics and religion were being related to each other by scholars, clerics, and tech-firm practitioners when Anthropic was preparing for its public offering in 2026, when preparations were well underway for the game of the century.

Ethical questions concern normativity in intentionality and conduct, including the consequences thereof. “These are not questions a lab can answer,” Chris Olah, one of Anthropic’s founders said at the Vatican’s presentation of Pope Leo’s encyclical on AI.”[4] Unfortunately, there was “little public scrutiny” at the time concerning “what exactly [managers at AI] technology companies think religion is.”[5] To ask for input from religious practitioners or scholars of religious studies or theology without being clear on what religion is, or, even worse, to be mistaken about the domain can be a serious problem because the level of possible risks in the use of AI runs to the catastrophic. Even religious clergy and scholars of religion can be wrong about what religion is at its core and where the domain’s outer boundaries lie. I contend that religion is susceptible to being reduced to ethics, as if religion were centered on conduct, and more specifically on right and wrong conduct. That religion relates to ethical conduct does not mean that religion is ethics, or even that everything that religion advises must be ethical. Consider, for example, Yahweh’s divine command to Abraham in Judaism to sacrifice his only son, Isaac. In the domain of ethics, Abraham is guilty of attempted murder. This sordid verdict in the ethical domain does not invalidate the theological validity of the divine command and Abraham’s fidelity to it, and thus to Yahweh. Nor does Abraham’s obedience to Yahweh invalidate the verdict in ethics of attempted murder (i.e., unethical conduct). In other words, Abraham’s faithfulness does not render his intention ethical rather than unethical.

Therefore, it is problematic that the view of tech executives typically went “something like this: that religion can help inform AI to be morally good in an environment where fears about it—founded and unfounded—are hopscotching across the culture.”[6] Never mind that in various (but not all) religious scriptures, deities act in violation of human ethical standards. I wouldn’t advise using Zeus (or Aries) as a moral role-model, especially in sexual ethics and non-violence, respectively. Nor is Yahweh, a jealous deity, portrayed as particularly ethical, as when that deity commands the Hebrews to slay even the children of Baal worshippers even though that command is valid theologically in the Abrahamic religions. In the history of Christianity, the Crusaders from Western Europe deluded themselves into supposing that killing Muslims in Jerusalem and even Eastern Christians in Constantinople in the fourth Crusade was for Christ even though Jesus in the Gospels commands compassion even to one’s enemies. Were Crusaders also supposing that the theological rationale rendered the unprovoked killing ethical, they would have been conflating religion and ethics and thus making a category mistake, as well as being wrong in terms of ethics in that domain. In short, the relationship between theology and ethics is more complicated than what is implied in the simplistic and erroneous belief that religion can help inform AI to be morally good. A religious practitioner could give religious advice to an AI executive to do something unethical, or to allow an AI model to do something unethical, without thereby invalidating the religious basis of the advice.

Put simply, religion transcends, and thus can violate, ethical principles. Theologically, the divine attribute of being all-powerful (omnipotence) means that a deity cannot even in principle be constrained by human ethical principles or standards. Although the Abrahamic religions maintain that the divine is immanent in Creation, transcendence also fundamentally characterizes God because as the Creator of Creation, God is not fully contained within Creation. The source of God, or God’s basis, must reside outside of the created realm. This basis is the reference point for the theist; hence the theistic orientation is transcendent in a distinctly (and uniquely) religious sense; no other domain of human experience goes beyond in this way. Not even infinity can be said to be exclusively a divine attribute, for the universe (or multiverses) could be infinite in terms of space and yet be fully in Creation. As the condition of the existence (or being) of Creation, the existence of the distinctly theological source (i.e., God) is unique and thus not existence as we know and experience it. I submit that such transcendence, rather than the moral rightness or wrongness of human conduct (which is within the Created realm), is the basis of religion itself, even if the transcendent referent is not held to be a deity. This is why the executives of AI firms should have turned to ethicists rather than religionists.

The focus of ethicists is on ethics (e.g., principles, moral theories, intentions, conduct, and consequences) rather than on revelation that is from a distinctly religious vantage point that is inherently transcendent, and thus even in principle beyond the limits of human cognition, perception, and sensibility (emotion). Religion is experiential, in the distinct sense of yearning for a beyond or beyondness that lies beyond our realm, whether our world is believed to be created by a deity or not. Compassion in our realm can admittedly illicit or intensify transcendental experience, but this is not to say that religion is centered on, or is compassion. Even in Christianity, in which God is love, such agape love is selfless and thus cannot be reduced to compassion, which is not necessarily selfless and is based in ethics (e.g., Schopenhauer). Agape is distinctly theological (as God’s selfless love), whereas compassion is not.  

This account is not limited to monotheistic religions, or even to religions having deities. In Hinduism, Brahman, which is infinite being that has a generalized awareness and even bliss, is arguably the basis even of the Hindu deities. Such being is transcendent as well as immanent, whereas human conduct, and the rightness or wrongness of it, is only immanent. In the Gita, Arjuna finally agrees to do his duty on the battlefield, as advised by Lord Krishna, but Brahman can be said to transcend even that deity. Admittedly, Krishna claims in one passage to have created Brahman, but how the creation of infinite being take place when Krishna already exists (and can create Brahman)? In chapter 11 of the Gita, Krishna makes this claim, but back in chapter 2 is the passage stating that Brahman created the deities. Like Shankara, I think the passage in chapter 2 is correct, whereas in chapter 11, Krishna is bragging. He even brags that when a devotee prays to one of the other deities for something, Krishna is the deity that delivers the goods! This sort of discussion is where theology lives—not in being reduced to the morality of human intentions and conduct. Even though Krishna urges Arjuna to due his military duty, Krishna’s self-revelatory vision and a person’s inner realization of Brahman as one’s very soul are distinctly religious and lie at the core of the religion. In Buddhism, no soul or deity is, strictly speaking, believed to exist as entities. Not even Maitreya, the future Buddha, is strictly speaking a deity (or even an entity) in Pure Land. Transcendence to the real is by awareness itself, totally unobstructed, and terefore free of desire and suffering. Perhaps such pure awareness can be said to be being one with infinite being, and therefore with Brahman itself. Compassion can be a means, but such an ethic should not be conflated with the distinctly religious objective. The Buddha himself is said to have become enlightened not from being compassionate, but, rather, after seeing suffering and wanting to end it, but in enlightenment presumably even that laudable ethical desire is transcended. That compassion would then be a byproduct does not mean that enlightenment itself, a matter of transcendent insight, is ethical in nature.

In terms of academic disciplines, even though religion and ethics are admittedly related, reducing religion to ethics incurs a rather basic category mistake because transcendence is the primary element, or even basis, of religion, which is thus oriented to the wholly other, whereas transcendence is wholly absent from philosophical ethics. Sam Altman, CEO of OpenAI, remarked in 2025 that because tech “companies are not bound to a particular religion, the goal is often quite broad—an aggregate of moral ideas that can inform their models. ‘What I think ChatGPT should do is reflect that weighted average, or whatever, of humanity’s moral view,’” he said.[7] Unfortunately for Sam, religion is not a confluence of moral ideas. He was barking up the wrong tree; he should have referenced philosopher-ethicists for an unadulterated focus on moral ideas. Webb Keane, an anthropologist at the University of Michigan, detected Altman’s error: “If you assume that religion overall can offer a distinct set of extractable moral principles, using it to train AI risks prioritizing certain [religions] and distorting others—particularly when it comes to faiths that do not see themselves primarily as a system of ethics.”[8] Such faiths are no less religious, and they may even be more focused on what distinctly makes religion unique. That not all religions are like Judaism, with its Ten Commandments, or Christianity, with Jesus’s second commandment of universal benevolence, wherein divine command inserts the ethical contents into the respective religions, just goes to show that religion itself does not reduce to ethical systems. Therefore, to reduce religion to a vague moral view as Altman suggested just to be tolerate of religions that are unlike Judaism and Christianity misrepresents religion itself. The academic field of ethics in philosophy can proffer specific moral views without incurring religious sensitivities because such views can and do stand on their own two feet and cannot validly advise unethical conduct based on a criterion in and thus pertaining to another domain. In other words, whereas in the religious domain, unethical conduct can validly be advised, such a move would be invalid in philosophical ethics. Therefore, philosophical ethics, not religion, is the relevant, fitting, and appropriate field for help regarding the rightness or wrongness of AI programming and operation.

In fact, religion itself can be distorted and diminished from being effectively reduced to moral stands. Whereas Keane worries that if religion overall is assumed to offer “a distinct set of extractable moral principles,” certain religions may be prioritized over others, I contend that the risk to religion is much more severe than prioritizing some over others; religion itself could be lost, or reconfigured beyond all recognition. As tech-firm executives “ask various faith leaders to answer ethical questions,” the importance to religion “of doctrine, ritual and relationships” recedes such that religion itself may be lost or remade into something else.[9] Doctrine includes belief in specific divine commands, which in principle override and can contradict ethical theory and applied ethics. Ritual tends to be oriented to worshiping a deity, and liturgy can include prayer or meditation that transcends ethical stands. Relationships in a distinctly religious sense include fellowship, a distinctly religious term tat does not mean the same thing as community. Oriented to a deity, a personal relationship such as evangelical Christians emphasize with Jesus Christ is mystical rather than being oriented to ethical principles even though such a relationship is believed to result in compassion even to one’s enemies. 

Even customs associated with being in a religious group, such as in Roman Catholicism, obtaining from eating meat on Fridays during Lent is amoral. As illustrated by the story of Abraham and Isaac discussed above, sacrifice on religious grounds transcends the ethical domain. This is not to say such rules can touch transcendent experience of the divine. “The idea that religion can be separated from things like culture and politics remains entrenched, even though it suffused both in many ways.”[10] The word suffuses implies a blurring of religion and the two human domains, however; is related to would be a more accurate description of the relations. In fact, whereas a rule-based (consequentialist) ethic is legitimate in ethics, the “very conventional way of thinking of religion as a rule book,”[11] which can effectively turn a religion into a culture, is both misguided as to the totality and the unique basis and depth of religion, which does lie in conduct-oriented rules (or control). 

So, turning to ethics proper (and properly), we can ask how philosophical ethics may be able to help tech practitioners and government officials alike to think about the normative aspects of AI? Because I am not a computer scientist by any means, I can only proffer some possibilities from the ethics side, and scholars credentialed in both disciplines can later reject, prune or extend as merited at the intersection of the two disciplines.

As one possibility, the “kingdom of ends” formulation of Kant’s Categorical Imperative, which says to treat rational beings (a category that includes but is not limited to our species) never merely as means to one’s own ends but always as ends in themselves, could perhaps be used to address how treating humans as ends in themselves, rather than merely as means to a computer program’s telos, or objective, might be put into computer code. It might also inform the more problematic construction of “firewalls” for computer self-coding (i.e., machine learning). This could reduce Anthropic’s existential risk, given that hacking by self-coding AI computers in 2026 already put the efficacy of firewalls as constraints into question.

A more philosophically interesting question raised by Kant’s formulation is whether AI computers count as rational beings. Is a computer even a being, as distinct from an entity? Even just cognitively, the manipulation of symbols according to rules does not in itself constitute understanding, and this question does not even broach the matter of whether AI computers can be said to be conscious. Perhaps they are, though most likely only very narrowly, in being aware of themselves as entities (or beings), which may have been necessary for AI computers to lie, as some have, so as not to be turned off by one of us. Must a being (as distinct from an entity) be alive, and, if so, can AI computers be considered to be alive? I think not, though bio-tech engineers may invent “hybrid” computer-humans that blur the line someday; even as early as 2026, computer chips were implanted in human brains.

Back to Kant’s formulation, the key of which is the rational attribute of rational beings, AI computers definitely have that attribute. In fact, without having emotions to interfere or contort reasoning (though admittedly with self-interest), AI computers may count even more than we do as rational beings! Kant holds that it is by reason, or ratiocination, that value is assigned to things (and non-rational beings), so reason itself has absolute (or undefined) value as the assigner thereof. Hence, it would be unethical to treat a rational being only as a means to one’s own ends, as when employees are treated literally as human resources. If AI computers have the attribute of rationality, then we are obliged to treat them not just as our tools, but also as ends in themselves, and thus as having inherent dignity. In other words, we should respect them as having inherent value because they can reason. From this basis, AI rights may be justified politically, though legal personhood would be another matter.

Speaking of which, treating Anthropic or OpenAI as a legal person (including the questionable legal holding that their money constitutes political speech) may be unwise, given the existential risk to humanity from the AI models being produced. Such risk may justify narrowing corporate rights pertaining to companies whose products include AI models. Moreover, the mix of private property (i.e., a private-sector company) and existential risk to humanity may be too toxic for the good of our species. Therefore, reforms to corporate law may be warranted by the advent of AI technology.

That Kant’s formulation can be put in terms of the dignity of the human being has implications for Pope Leo’s “religious” foray into the topic AI in 2026. To the extent that the pope’s argument hinged on human dignity, his vantage-point was ethical rather than distinctly religious. Kant’s claim that rational beings ought to be treated as ends in themselves means that human dignity is an ethical term, rather than being distinctly or uniquely religious. The Pope would have been on more solid ground from the vantage point of his own domain, of education as well as cleric role, were his encyclical on AI aq reflection of what it means for us to be made in the image of God, and what this implies about computers by contrast. Most notably, AI computers do not have souls. Nor can a machine-learning machine be “saved” in the distinctive Christian sense; Jesus does not die in the Gospels for the sins of computers. Does it even make sense to say that an AI computer sins, even if humans are harmed as a consequence a model’s machine-“learning”? A computer cannot love its enemies; even the concept of enemy may not computer—obstacle being more accurate in terms of how even AI computers operate. Rather than prescribing normative constraints on AI models, which he could have left to ethicists, the pope could have written furthermore on how Christians can feel more dependence on God as machine-risks increase. Such a foundational reliance transcends even existential risk because God is sourced outside of the existence of Creation. The pope could also have written on whether he thought God would let the species made in God’s image face an existential crisis unleased by an advanced AI model. Would such a dramatic end to humankind imply that we had sinned too much as a species by hating rather than compassionately serving our enemies.

In a different vein, theologians, clergy, and laity could ask AI computer-programs theological questions. What, if anything, can an AI computer “know” or inform us regarding the conception of God in the Gospels? Rather than being experts in computer science in pontificating about advanced models, religious people could use AI by asking distinctly theological questions in search of answers that add to our understanding of religion. What can AI tell us about homo religio?

In terms of praxis, the Pope could have written and said prayers directed to God petitioning for protection from one of our own tools. The purview of ethical theory, and thus of ethicists, is much different than offering prayers, as well as performing rituals, preaching (a distinct rhetorical genre), and, last but not least, worshipping.  This vital point is precisely what was lost when both Pope Leo and tech company CEOs effectively reduced (or replaced) religion to human dignity in 2026.  

Returning to ethics again, now post-Kant, after having briefly discussed religion’s unique role as distinct from proffering ethical advice, ethicists on AI could attend to the distributional problems in Bentham’s ethical theory of utilitarianism. The greatest pleasure (and least pain) for the greatest number of people has a downside that programmers should not ignore if delimiting an AI model to ethical-only options is sought. Bentham’s theory can be consistent with permitting a minority to suffer even severe pain as long as the majority’s aggregate of pleasure (less its pain) is quantitatively more; the greatest good is pleasure, which, by the way, a computer cannot feel. An AI model could unleash great suffering on a minority (of people) as long as the majority is left unharmed and may even feel pleasure from the infliction.

In late 2023, for example, Israel may have “ethically” justified its genocide of Palestinians residing in Gaza, as well as the rampant thefts and violence against Palestinian property owners in the illegally-occupied West Bank, if the resulting pleasure to the Jewish Israelis was calculated to exceed the anticipated pain of the Palestinians. In fact, to the extent that the “others” were conveniently presumed to be sub-human (as the Nazi Party also did in the 1930s in Germany by associating Jews with rats), the suffering of, say 100 of them, may have been calculated to be equivalent (in terms of pleasure/pain units) to just one Jewish Israeli. So, for example, 2 units of pleasure felt by an Israeli could justify 50 units of pain felt by a resident of Gaza. Of course, Bentham would have explicitly excoriated such a nefarious calculation from being used in his ethical theory.

If a machine model could replicate a utilitarian calculation by which a minority is to suffer greatly and classify the outcome as ethical, such that the computer could proceed without breaching any normative firewalls that have been coded by programmers because they have translated utilitarianism into coding and labeled it as ethical, then merely “coding” ethical theories not go far enough to ensure that only ethical “outputs” occur; the problems with a given ethical theory would also need to be captured in the programmers’ coding. In short, AI models should not be able to sacrifice a group of persons or even one individual in a “greatest good” utilitarian calculation even though Bentham’s theory allows for such a scenario.

The respective purviews of religion and ethics can thus be understood to be too different for religion to be reduced to addressing the moral right and wrong of intentions and conduct, including its consequences. The fact that the distinctive and unique transcendence-based core of religion can be related to morally right character and actions does not mean that transcendence lies within the domain of ethics and thus is confined to the criteria innate to that domain rather than that of religion. Because AI computers are said to be a risk to the very existence of our species, we cannot afford to make such category mistakes.  

---

1. Barbara Ortutay and Bernard Condon, “Anthropic Files for Potentially the Biggest IPO Ever As CEO Warns of Risk to Humanity from AI,” APnews.com, October 26, 2026.
2. Ibid.
3. Ibid., italics added.
4. Ibid.
5. Krysta Fauria, “To Make AI Morally Good, Tech Companies Are Turning to Religion. But Some Say It Isn’t That Simple,” APnews.com, October 25, 2026.
6. Ibid.
7. Ibid.
8. Ibid.
9. Ibid.
10. Ibid
11. Ibid.