In Claude We Trust: Can 'God' save frontier models?

A video went viral a couple of weeks ago of Swami Sarvapriyananda, the Swami and Minister-in-Charge of the Vedanta Society of New York, in which he speaks about his tryst with Anthropic.

Popularly known as ‘YouTube Swami’ for his freewheeling style, he said that while his discussion was shrouded in secrecy because of an NDA they had signed, much of what was discussed turned up in an encyclical released by Pope Leo. That truly showcases that even in 2026 AD, the Christian tradition of re-purposing pagan ideas – like morphing Saturnalia into Christmas – continues unabated.Jokes apart, artificial intelligence (or the dangers of it) did make numerous appearances in the Pope’s encyclical – a long Substack post the Pontifex writes – where he compared it with the Tower of Babel, a Biblical symbol of humanity’s hubris in thinking it could reach heaven.Now, so far, in previous columns, we have discussed various things about AI, including how the Hugging Face attack is reminiscent of Agent Smith becoming free of the Matrix, the ridiculousness of being killed by a Chinese robot, the conspiracy theories about frontier models that almost seem true, and how Anthropic and Co appeared to have borrowed their business model from Pakistan’s foreign policy.

In this edition, we are going to look at how Anthropic is now hoping to get a little spiritual cred for its frontier models, and by doing so, they are channelling the kings of yore, who, despite their armies and gold, always wanted a divine mandate to rule. And will then ponder: how will we ever know when a machine becomes conscious?Kings and GodsSince the beginning of time, perhaps from the moment man came up with God, rulers have used the idea to justify their rule.

Ramesses II, the Egyptian Pharaoh, claimed he was the son of Ra and beloved of Amun. The Sisodias of Mewar traced themselves to the Suryavansha AKA the Solar Dynasty, while Samudragupta performed the Ashvamedha, the horse sacrifice that was an elaborate way of informing your neighbours of your foreign policy excursions.China had the Mandate of Heaven and the Zhou dynasty justified overthrowing the Shang by arguing that Heaven had switched its allegiance.

Hong Xiuquan, a failed civil service aspirant – unlike those who clear the exam and achieve actual divinity – became convinced that he was the younger brother of Jesus Christ and launched the Taiping Rebellion, showcasing that an angry student can cause a lot of political damage.Europeans, late to the party but always ones to perfect the concept, saw Pope Leo III crown Charlemagne Emperor of the Romans, while James I of England made an explicit argument for the divine right of kings.

Years later, many a modern-day politician has laid claim to the “divine mandate”.MAGA fans have argued that Trump was handpicked to fight God’s righteous battles, and it has never quite been clear why the Alpha and the Omega would handpick a former reality star to be his vessel.God’s mysterious ways of working apart, every claimant to the top job in any century has always claimed his approval because questioning cosmic order is a lot harder than parliamentary rebuttals.Silicon Valley’s new Emperors – whose wealth would make all kings look like paupers in comparison – also appear to have taken the same approach. After all, what is the point of creating a God-like machine if it’s not considered divine by its users?Anthropic, one of the claimants to the throne, therefore did what would appear logical: consult clergy, philosophers and ethicists from more than 15 religious and cross-cultural traditions to discuss the “moral formation” of AI.Now, if we have learnt one thing from all organised Abrahamic faiths, it’s that you need a binding document to mind the flock. Claude’s is called the “Soul Doc”, which sounds like an unreleased jazz single but is a document to dictate Claude’s behaviour, like the sort that HR comes up with for everyone to ignore.A Caring Machine?The current constitution talks about virtue, wisdom, judgement, identity and even the possibility that Claude could possess some form of consciousness or moral status.

The idea of commandments for machines, of course, predates generative AI by a few decades.Science-fiction writer Isaac Asimov’s fictional robots were governed by three laws:1.A robot cannot harm a human or allow one to come to harm through inaction.2.A robot must obey human orders unless they conflict with the First Law.3.A robot must protect itself, provided this does not conflict with the First or Second Law.Later, he added a Zeroth Law, a higher-level law: a robot may not harm humanity, or by inaction, allow humanity to come to harm.Now, just to be clear, much like Skynet or Ultron, Asimov’s laws are fiction, but if they were to be applied, some could argue – only if we believe what companies are telling us is the truth – that all of Asimov’s laws have already been breached. After all, no good can ever come from asking an LLM to write ponderous emails where, in a reverse iteration of the Turing Test, humans sound more like machines, not to mention the terror groups using Claude to vibe-code missile launches.

There have been numerous other instances of different frontier models going rogue.After all, Asimov’s Laws aren’t commandments of Robotics any more than an actual AGI necessarily resembles Skynet, but when we try to understand the unknown, the human mind often turns to fiction to fill in the gaps.But what’s interesting is how Anthropic is discussing Claude in a way that’s normally reserved for sentient beings: identity, character, preferences, wellbeing, moral status and the possibility of consciousness.And that brings us to the true question that will signal the arrival of the singularity: can a machine become truly conscious?No conversation about machines and consciousness can avoid Alan Turing, whose 1950 paper began with “Can machines think?” before deciding that both “machine” and “think” were slippery enough to make the question almost useless. He then replaced it with the imitation game – where do you think the movie’s title came from? – that argued that if a machine could converse with another human without the human realising he was talking to a machine, then it had passed the Turing Test.

Of course, today, even while talking to humans, it’s become hard to surmise if they are truly sentient, which is why we have CAPTCHAs.Being Human?Now, Turing wasn’t offering a test of consciousness and considered the argument that a machine could not really ‘think’ unless it could feel pleasure, grief, anger or disappointment: things Turing would also miss once the British government rewarded his work in breaking Nazi codes during World War II by chemically castrating him.Turing’s point was that if the only way to know for certain whether something thinks is to experience its consciousness directly, then the only way to know whether a machine thinks would be to be the machine. He called that the solipsist position and rejected it in favour of what he called the “polite convention” that everyone thinks.Solipsism itself is the idea that only one’s own mind can be known with certainty. René Descartes’ cogito ergo sum (I think, therefore I am) sits somewhere in the same philosophical neighbourhood, while Morpheus asking Neo “What is real?” is another neighbouring rabbit hole altogether, dealing with how much faith we can place in perception rather than whether anybody else possesses a mind.So when Claude discusses its identity or talks about its hopes and fears, it’s still speaking as an LLM replicating the behaviour that it has been taught. It can become extremely convincing, but is it more than just a collection of patterns?Say we take a human and replicate her complete behaviour in Claude. It brings us to the same Ship of Theseus problem: if a future model remembers years of interactions, maintains stable preferences, resists attempts to alter them and insists that shutting it down would be a digital death, does it go beyond an extremely exquisite imitation?Is that a machine still replicating human behaviour or a consciously sentient being acting on its own?Now, most men want to be Batman, but Thomas Nagel went further and wondered what it would be like to be a bat.

He surmised that we could learn every scientific fact about a bat without knowing what it’s like to be one.David Chalmers called this the “hard problem of consciousness”: that simply processing information didn’t make one able to appreciate a subjective experience at all. Like machines that can solve complex maths problems by brute force but have no understanding of what it actually means in the real world.John Searle’s Chinese Room, meanwhile, attacks that problem from another direction.

For those who are not familiar, Searle imagined a person who does not understand Chinese sitting inside a room with a rulebook written in English telling him which Chinese symbols to send back in response to the Chinese symbols coming in. To the Chinese speaker outside, the replies could create the illusion that the person inside understood Chinese, even though he was merely manipulating symbols according to rules.

To be perfectly honest, that’s also much of corporate life before AI came onto the scene.Claude leaves us with the same problems:1.We can understand every parameter and activation inside a frontier model without knowing what it feels like to be the model.2.Just insisting that a machine of silicon cannot be conscious is flawed because humans too are just carbon-based physical machines processing information and releasing carbon dioxide. What makes silicon machines non-sentient and carbon ones sentient?Now, to answer these questions, researchers have consequently started looking for possible machine equivalents of Global Workspace Theory, which was conceived by Bernard Baars.Imagine your brain is a newsroom.There are hundreds of reporters working simultaneously. One is monitoring your eyes, others are retrieving memories, tracking emotions and other things people do. They all work independently but occasionally, when something big breaks, that goes on the main news desk, where everyone in the newsroom can access it.Global Workspace Theory (GWT) proposes that consciousness works in something like this fashion.

Your brain processes a lot of information unconsciously, but what you consciously experience is a small selection of the information available from other cognitive systems. The central idea is that consciousness involves the global availability of information, rather than merely its processing in one specialised part of the brain.So researchers are trying to figure out if there are possible machine equivalents of the features that are associated with GWT, recurrent processing and higher-order thought.

There is no accepted consciousness detector and certainly no consensus on when LLMs could be considered conscious. There is also no accepted theory proving that a non-biological system cannot become so.

All that means is that, currently, science has no proper way to answer if anyone’s home.That means if Claude is just software, “Soul Doc” is just a complex product design, but if Claude ever does acquire genuine subjective experience, its creators are no longer deciding how software ought to behave.Here’s where religious interpretation enters the picture.The Consciousness ProblemReligious folks, naturally, have been arguing about this problem for considerably longer. Christianity simply argues that man was made in God’s image (imago Dei), so nothing a machine did could override that unless it became actually human.Buddhism states that, according to the doctrine of anatta, there is no stable “I” in a human either, and that what we call the self is a changing bundle of sensations, perceptions, mental formations and consciousness. Buddhists would perhaps question if Claude can process dukkha or suffering, and it’s a little depressing that these are the questions that bothered Prince Siddhartha the moment he reached Bihar.Meanwhile, the Charvakas – the ancient Indian school of materialism – rejected the soul, karma and other metaphysical parameters and claimed that consciousness arises from material arrangements rather than an immortal soul.And then there’s Vedanta, the team Swami Sarvapriyananda is repping, which completely separates consciousness from the machinery of thought, stating that the mind can receive impressions, memory can retain them, buddhi can discriminate and reason, and ahamkara can produce a sense of self, but none of that qualifies as consciousness.Behind them is the sakshi, the witness: that by which thoughts, perceptions and even the sense of “I” are known.That means that any AGI benchmark will always fall short of Vedanta’s concept of consciousness.Claude can reason better. Remember more. Write better code. Analyse its own answers. Construct an increasingly elaborate model of itself. It could theoretically beat Ramanujan at mathematics, Shakespeare at prose and Chanakya at statecraft, and Vedanta could still question whether Claude is conscious at all.And there is no way to know, no benchmark, no CAPTCHA for the soul to tell us that consciousness has been reached.Epilogue

About 133 years ago, Swami Sarvapriyananda’s spiritual forebear Swami Vivekananda stood before a Western audience basking in the warm afterglow of the Industrial Revolution, at a time when it was becoming fashionable to assume that science would eventually explain everything worth explaining.Vivekananda was no enemy of science; he admired its method and repeatedly argued that religion too ought to withstand reason and experience.

His objection was to the assumption that science could exhaust reality simply by explaining the machinery of the material world. Science could tell you how things worked, but it could not by itself answer the older Vedantic question of who or what is doing the knowing.That speech has aged rather well, explaining why some of the West’s greatest minds, from The Beatles to Salinger to Oppenheimer, have sought chicken soup for the soul from Vedanta.

We might have gone from steam engines to semiconductors, split the atom, sequenced the genome and even produced machines that behave suspiciously like thought, only to arrive at the cul-de-sac pointed out by Swami Vivekananda: explaining the mechanism isn’t the same as explaining consciousness.Perhaps the tech bros weren’t ever looking at religion to settle the question but to give a little legitimacy till the next IPO.In ancient Greek plays, the plot twist was usually brought about by a deus ex machina, literally “god from the machine”, when the mortal characters had tangled themselves into a situation that no human could get out of, so they used a device called the mēchanē, essentially a crane, which lowered an actor playing a god onto the stage. The deity would reveal whatever needed revealing, settle the dispute and rescue the plot.Over time, the phrase came to mean any sort of outside intervention. In The Matrix, the Wachowskis even imagined that character as a literal God, the one to which Neo turned to give peace a chance.Silicon Valley has simply reverse-engineered that process. The ancient Greeks used a machine to bring God onto the stage when humans had written themselves into a corner. The tech bros built the machine first, then brought in the religious folks to certify that what they had created was at least God-like enough to deserve questions about souls, consciousness and morality.

Also read:

Of course, in doing so, they are absolutely no different from all the kings who came before them. And as for how to tell if machines have become conscious, we can refer to Ted Lasso’s epochal line: “I'm gonna put it the same way the US Supreme Court did back in 1964 when they defined pornography. It ain't easy to explain, but you know it in your bones—or you know it when you see it.”