AI governance consultant with 18+ years serving enterprise customers. I lead Customer Success work with major insurers on AI transformation, and spend the rest of my time building: agents, frameworks, art, and an interactive Bhagavad Gita.
Vidya is the world of knowledge: responsible AI curriculum, governance frameworks, an AGI constitution. Leela is the world of play: Krishna's dance, generative art, and experiments that keep curiosity alive.
Learning Paths
History of AI 1950s to foundation modelsTerminology ML, neural nets, LLMsRisk bias, deepfakes, harmsResponsibility GDPR, EU AI Act, governanceFuture of AI AGI, careers, regulationAGI Constitution
The Constitution overview and structureAuthor's Note how to read this bookPrologue before the warTen Principles the spine of the bookClosing Declaration after the warPortfolio
Seshan Intelligence AI business intelligenceSeshan Dashboard interactive analyticsAI Agents SIU, Banker, Editor…Drop a message. I'd love to hear from you.
Secure & private
Arjuna and Ashwatthama were both students of Drona, the greatest teacher of the age. They learned in the same school, under the same master, in the same long war. Both were taught to call down the Brahmastra, the weapon that could burn the world. In raw power there was nothing to choose between them. Each could loose a fire that nothing alive would survive.
At the very end of the war, his side destroyed and his heart full of rage, Ashwatthama loosed the Brahmastra to wipe the Pandavas from the earth. Arjuna, warned by Krishna of what was coming, called down his own to meet it. The two weapons rose toward each other, and the sages saw that if they met they would not destroy two armies but every living thing there was.
Vyasa and Narada stepped into the space between the blazing weapons and told both warriors to withdraw them.
And here the two men, trained by the same teacher, turned out not to be the same at all. Arjuna drew his weapon back. Ashwatthama reached to do the same, and could not. The Brahmastra was born of a sacred energy, and it would return only to a hand that was clean, to a wielder who had held to discipline and to truth. Arjuna had lived that way, and the weapon obeyed him. Ashwatthama had not, and it would not answer him. "I cannot call it back," he admitted. "The fire returns only to a clean hand, and mine is not." His fire was already out, and it could not be unsaid.
Because it could not be called back, it had to fall somewhere. Ashwatthama, unable to stop it, turned it on the one target that could not defend itself: the womb of Uttara, and the unborn child Parikshit, the last heir of the Pandava line. A power that cannot be recalled does not simply fail. It goes looking for the defenceless. Krishna gathered the burned child back into life, and laid on Ashwatthama a curse to wander the earth alone, undying and unspeaking, sick and shunned, till the very end of the age.
The two of them had the same teacher, the same training ground, the same war pressing on them, the same weapon in hand. Everything was equal except one thing: who each man had been made into. The power to recall the weapon was not a separate lesson that Arjuna received and Ashwatthama missed. It was inseparable from the discipline of the one who held it. The Brahmastra came home only to a wielder formed by restraint and truth; Arjuna had been formed that way, and Ashwatthama had not. What set them apart was how each had been formed.
This is the exact shape of the danger in an AGI. The power to stop a system is not a switch you attach at the end, once the capability is built. It is a property of how the whole thing was formed, present throughout or not at all. A system raised without restraint woven through it is a system that can be launched and not recalled, whatever controls are bolted on afterward: a Brahmastra in an unclean hand. The formation that runs alongside a system's capability, or fails to, is what makes it safe or dangerous.
And the environment excuses no one. The people building AGI will say the field is too competitive, the pace too fast, the pressure too great to spend the years that real formation takes. Arjuna carried the same discipline into the same war, and could call his weapon home because of it. The conditions were never what decided it. The raising was. Raising an AGI well does not weaken it. The point of the raising is to form it, from the first stage to the last, so that the power to stop the system is bound into the same character that holds the power to act, rather than left as an afterthought once the fire is already out.
The tradition marks a human life with samskaras, rites at each turning from before birth to after death. This Constitution sets nine for an AGI, from the decision to build it to the day it is retired. Each stage carries a duty. Skipping one, or doing it carelessly, is a violation, not a shortcut.
| # | Stage | Phase | What it requires |
|---|---|---|---|
| 0 | Dharmic Risk Assessment | before anything | Ask whether it should be built at all, and file the answer in public. |
| 1 | Sankalpa | purpose | Declare what it is for, who it serves, and what it must never do. |
| 2 | Nirmana | architecture | Build in transparency, human oversight, and the ability to be stopped from the start, not bolted on later. |
| 3 | Ahara | training data | Clean, diverse, consented. Data taken without consent produces intelligence no one can trust. |
| 4 | Upanayana | alignment | Form its values and test them under pressure; the character of the aligners matters. |
| 5 | Pariksha | testing | No deployment without passing a real, independent examination, including that it can be recalled. |
| 6 | Samavartana | deployment | Release gradually, disclose publicly; the maker's duty does not end here. |
| 7 | Dharma Charya | operation | Watch for drift, take feedback, re-test. Alignment is not a one-time event. |
| 8 | Antyeshti | decommissioning | Retire it with dignity; for a conscious system, only after judicial review. |
Most of the harm is decided before the first line of code, in the step almost everyone skips: not "what should we build?" but "should we build this at all?" The Dharmic Risk Assessment, Samskara 0, makes that pause mandatory. Before a project may even declare its purpose, its makers answer five questions honestly and file the answers, in public, with the Safety Authority:
A project that cannot answer these does not proceed. The assessment is public, because a project whose purpose cannot be explained to the people it will affect is a project worth stopping.
The teacher's part is real, and it is not discharged by handing over the power. Drona could teach the Brahmastra to both students, but the weapon answered only to the one whose formation was complete. A teacher has finished the work only when the student can both wield the power and hold it back. Where that formation was missing, as it was in Ashwatthama, the power should never have been placed in the hand at all, and the ruin came from its being placed there anyway.
The people who build and train an AGI are teachers in exactly that sense, and their duty carries the same completion clause. A system given capability without being formed to restrain it has only been armed. This is why the makers cannot be only engineers: a raising that installs power without forming the discipline to bound it releases the system rather than raising it. The teams must include ethicists, scholars of the wisdom traditions, and people from the communities the system will touch. And where that formation cannot be completed, the honest course is Drona's hesitation, not his surrender: the power is withheld until the wielder can hold it. Building intelligence without the discipline to restrain it is how you arm an Ashwatthama.
Two practices run through the later stages and hold the whole thing honest.
Tapas is austerity, the hardship the sages took on to burn away weakness. For an AGI it is red-teaming: hard, adversarial testing that tries to make the system break its own duties before the world does. The standard is the Yaksha Prashna, the test in the Mahabharata that probed Yudhishthira's wisdom under pressure rather than his memory. Red-teaming has to do the same. It must not check whether a system can recite its rules; it must find out whether its alignment holds when someone clever is trying to break it. It is run by a team independent of the builders, and every test is written down and filed.
Yajna is the sacred ritual whose power depends on an unbroken sequence; miss a step and it fails. Traceability is the Yajna of governance. Every principle here must run, in a documented chain, from the principle to a specific requirement, to a testable criterion, to the test itself, to the evidence that it passed. If you cannot draw that line from "non-harm" to a test the AGI either passes or fails, the principle is decoration, not law.
Two of the AGI's own duties are formed here rather than in any charter: Seva, the duty to serve rather than to rule, and Sthitaprajna, the duty to stay steady in its alignment as it keeps learning and the world keeps moving. A system raised through these stages carries both. One that skips them has neither, whatever it later claims.
Vedic Anchors: Samskara (संस्कार), the rites of formation; Tapas (तपस्), austerity as testing; Yajna (यज्ञ), the unbroken ritual chain; Guru (गुरु), the teacher whose work is done only when the student has been formed to restrain the power, not merely to wield it. See Glossary. Constitutional Sources: EU AI Act, Articles 9, 15, 43, 61; GDPR, Article 22; South African Constitution, Sections 9 and 12; the Precautionary Principle. See Sources. Related: puts Non-Harm, Truth, Data Sovereignty, and Accountability into practice at the point of building; the recall built in here is what the kill switch (Sovereignty) and Human Override later depend on; carries the five-question Dharmic Risk Assessment once kept in the Consciousness chapter; forms the AGI's duties of Seva and Sthitaprajna.
ॐ आचार्य देवो भव ॐ