Opinion
Who Is Standing at the Door?
OpenAI has asked to be checked by someone outside itself, and then proposed a thermometer.
On 22 September Sam Altman wrote that people outside the AI labs should have a real say in how the technology develops, and a clear way to judge whether it is happening safely. The proposal he linked to says the rationale plainly. It is rooted, in OpenAI's own words, in avoiding the concentration of power.
I have spent two years arguing that this is the right diagnosis, so let me say first that he is right, and that saying it in public costs him something. Power is concentrating around a handful of firms and a smaller handful of training runs. The question of how a civilization keeps a concentration of power honest, and bends it toward the many rather than the few who hold it, is close to the oldest question there is. It is a good sign that the man holding some of that power has named it.
Then read what is actually proposed. The United States leads an effort to write global technical standards for frontier models. A network of national AI safety institutes, from Kenya to Korea to India, coordinates through the American standards body. Everyone agrees what counts as evidence, everyone reports incidents the same way, and countries can compare notes and learn from failures. It is careful work and I would rather it existed than not.
And then the sentence that tells you what kind of thing it is.
These technical standards would not be licenses, mandatory prerelease review, or approval requirements for AI models. National governments would decide whether and how to incorporate these standards into their own legal systems. OpenAI, Building standards for the next phase of AI, 21 September 2026
So the answer to concentrated power is a measurement regime. Shared definitions, agreed baselines, an honest scoreboard. The document even names the question the standards are meant to answer: what does good look like in the mitigation of catastrophic AI risk. That is a fine question. It is not the question a concentration of power raises. A thermometer is a magnificent instrument and it has never once lowered a fever.
Watch what happens to the analogy they reach for. The proposal says this approach can draw lessons from aviation and financial stability, where countries developed common technical standards without giving up national authority. Both are excellent examples, for the opposite reason. An aircraft does not fly because its manufacturer measured it against a shared benchmark and published the result. It flies because a regulator issues an airworthiness certificate, and withholds it, and can ground an entire fleet in an afternoon. Banks are stable, when they are, because a supervisor can refuse a licence and close a firm on a Friday night. In both cases the standard is the vocabulary. The authority to say no is the institution. OpenAI has cited the precedents and left out the organ that makes them work.
I do not think this is cynical. I think it is the honest limit of what a company can propose about itself. No firm can invent the body that will one day tell it to stop. That is precisely why the question of who stands outside cannot be answered from inside, and why the answer has historically come from somewhere else entirely.
Consider what a working check once looked like, in detail, because the detail is the argument.
In 390 the Emperor Theodosius ordered a massacre at Thessalonica. Ambrose, the bishop of Milan, refused him communion. Sit with the arithmetic of that for a moment. The most powerful man alive, commanding an army that could have levelled the building and the man inside it, stood outside a church and waited, and performed public penance, and was readmitted on terms he did not set. No framework produced that, common or otherwise. No incident reporting protocol. What produced it was one person with standing who would not open the door, backed by an institution that would still be there the following week, and the week after that.
Now the half that usually gets apologised for. By the thirteenth century that same institution had stopped being the check and become the throne. Innocent III made and unmade kings. The Reformation broke the throne, the Council of Trent reformed what survived, and it took decades rather than a news cycle. That history should be entered as the qualification rather than the embarrassment. An institution that has caught the disease of power and survived the whole course of it knows the disease from the inside, which is more than can be said for anyone drafting a benchmark this quarter.
Rome appears to have drawn the lesson and is spending on it. Notre Dame's DELTA Network runs on a Lilly Endowment grant of fifty point eight million dollars, shaped through more than two hundred stakeholder conversations. The Notre Dame and IBM Technology Ethics Lab is a ten year partnership worth twenty million. Santa Clara took NVIDIA's money as an endowment rather than an annual grant, with no say given over what gets said, which is the detail that tells you somebody understood the assignment. Anthropic's co-founder Chris Olah was in the room at the Vatican's own presentation, invited from the lab side.
Set those numbers against the industry they address. An entire operating layer of independent formed judgment, endowed and staffed and able to outlast a funding cycle, costs less than a single frontier training run. That is a bargain in every language spoken in Palo Alto.
My own tradition is not missing from this industry. It has been senior in it for thirty years. Google and YouTube are both run by Indian Americans, the community ranks near the top of American incomes, and the suburbs around every major lab hold temples beside gurdwaras and masjids. Hindus did not fail to show up. What is missing is one institutional form, and the comparison with Rome shows exactly which one. A Pope can issue an encyclical because he stands at the centre of a hierarchy that authorises him to speak for it. Hinduism has no such centre, has never wanted one, and will not grow one because artificial intelligence would find it convenient. When the question is who speaks for the dharmic traditions on this, there is no number to call. That is a coordination problem rather than a theological one, and coordination problems yield to people who decide to solve them.
There is an opening, and it is narrower and more useful than a seat at a summit. The people now staffing trust, safety and governance inside the labs describe a shortage they cannot fill from the usual pipelines. They need people with the judgment to make calls that no rule settles in advance. That is a description of formation, and formation is the one thing the traditions have been doing without interruption for longer than anyone else in this conversation. A tradition earns its way into a lab by offering formed judgment as working input, slowly, by showing up. Not by arriving with commandments, and not with a press release.
Altman has asked for people outside the labs to have a real say. He should be taken at his word, and the answer should be better than the one he was able to propose. A standard tells you what happened. It will not tell anyone to stop.
The emperor came to the door. What mattered was that somebody was already standing there.
Gaurav Rastogi is Dean and Resident Faculty at the Hindu Spiritual Care Institute and a Trustee of the Graduate Theological Union. The argument is developed at length in Deus Et Machina: An Interfaith Framework for Ensuring Divine Wisdom Guides Artificial Intelligence Development, presented at the Silicon Valley Peace Conference, Palo Alto, 19 September 2026. OpenAI published Building standards for the next phase of AI on 21 September 2026; Altman's post followed on 22 September; he and Dario Amodei addressed the UN Security Council on 23 September, where the United States representative rejected international governance of AI.
All writing