GAJRA Earth
Somebody has to keep the scoreboard.
A site that asks people to help define alignment owes them a map of who else is already arguing about it. This is that map: the bodies with authority, the promises the labs have made to themselves, and the camps that disagree, laid out so you can see the shape of the room before you speak in it.
This page is wrong already.
Last touched . For an actual measure of how fast this field moves, see Dr Alan D. Thompson's Life Architect, which tracks model releases and capability over time. This page was assembled by a system with a training horizon, so the recent end is exactly what it is worst at. Check every entry before you rely on it.
If you know this landscape better than the page does, which is likely, say so in public. A wrong date, a body since renamed, an instrument now in force, a camp described unfairly: all of it is a pull request, and the page is better for every one.
Two clocks, and one of them is personal
Why this project keeps organising itself around 2032, and why an Olympic deadline is the smaller of the two clocks, is set out on the destination page.
The honest position on both clocks is the same, and it is not neutrality. The opportunities are large. Plenty of people are unhappy about them, and on every side of both arguments there are points worth taking seriously. Compromise is not a risk to be managed here. It is what the outcome is going to be made of. What is unsettled is who steers, and how rules get navigated while nation states and bureaucracies turn slowly, well-funded companies turn fast, and a solo operator can change course in a day.
All of them are playing a gambit, in the proper sense: a move made early, on purpose, before anyone knows how the game ends. This page exists so that anyone making one here can at least see the board.
Who has actual authority, and what they have signed.
Binding law, treaty instruments and formal declarations, roughly in order of how much teeth they have. Dates and status need checking at source.
The EU AI Act
The first broad statutory regime for artificial intelligence, risk-tiered, with obligations phasing in over several years rather than landing at once. Whatever anyone thinks of it, it is the instrument other jurisdictions now write against.
Council of Europe Framework Convention
An international treaty on artificial intelligence, human rights, democracy and the rule of law, open to non-European signatories. A treaty carries a different kind of weight to a declaration.
United Nations
The Global Digital Compact and the advisory body work that fed it. Weak enforcement, wide membership, and the only forum where the countries with no frontier labs get a seat at all, which is precisely why it matters here.
The OECD AI Principles
Older than most of this and quietly load-bearing: the definitions and the risk vocabulary that a lot of later regulation borrowed from.
The summit series
Bletchley, then Seoul, then Paris, and successors since. Declarations rather than law, but they are where states put a position on the record and where the frontier safety commitments were made.
National safety institutes
The UK and US bodies and their counterparts, several renamed since founding, which is itself a signal about which way the political wind was blowing. They do evaluations rather than rulemaking.
What the labs have promised themselves
Every frontier lab now publishes a safety framework: capability thresholds that are supposed to trigger stronger safeguards, and commitments about what they will not deploy. Anthropic, OpenAI and Google DeepMind each have one under their own name, and there is a shared set of frontier safety commitments made at the summits.
Worth reading them for what they actually are. These are self-set thresholds, self-assessed, with the company deciding when its own line has been crossed. That is not nothing: a written commitment is easier to hold somebody to than a vibe, and several of them are detailed. It is also not law, and the people who wrote them would mostly say so.
This is responsible in the exact sense that page means it: a hand that stays on the thing, and a name that answers. The open question is whether a bound drawn by the party being bound can hold. That is question five.
Older mediators, same shape
The Vatican published Antiqua et Nova, a note on the relationship between artificial intelligence and human intelligence, and it is more careful than a lot of secular commentary: concerned with what intelligence is for and what it means to treat a person as a system, rather than with capability benchmarks. Other traditions have been at this too.
It belongs here for a reason that cuts both ways, though. These are institutions built on mediation. A small number of people interpret a higher authority for everyone else and ask to be trusted on the reading. That is close to structurally identical to what a frontier lab does when it publishes a safety framework: a few people, holding something the rest of us cannot inspect, telling us what it is and what they intend to do about it. Some of them get called gods by people who do not entirely mean it as a joke.
So the same test applies to both, and it is the one this site keeps applying to itself. A tradition that has spent centuries asking what a good life requires has real practice at the question. It has also spent centuries answering it on behalf of people who were never asked, and defending the answer with more than argument. Long experience is worth something. It is not the same as standing.
A company doing large-scale surveillance analytics named itself after the seeing-stones in Tolkien: objects that let a distant power watch everything and quietly bend whoever looked into them. Either unusual self-awareness, or nobody finished the book.
Who is arguing, and about what.
Described as fairly as I can manage. Anyone who finds their own position misrepresented here should correct it, and the correction stands as written rather than being edited toward the house view.
Catastrophic risk
Holds that sufficiently capable systems could cause irreversible harm, so capability should be gated on demonstrated control. Accused by others of speculative fiction that conveniently flatters the labs by implying their products are world-altering.
Present harms
Holds that the damage is already here and measurable: labour, surveillance, discrimination, the conditions of data workers, environmental cost. Accused by others of missing the scale of what is coming. Frequently sharper on the previous camp than on the labs.
Acceleration
Holds that delay has a body count nobody adds up: the road deaths a better-than-human driver would prevent, the illness that cheap preventative medicine would catch. Harms from moving are counted and named, harms from waiting are neither, which makes the comparison uneven before it starts. Also argues that heavy safety framing entrenches whoever is already ahead.
Open weights
Holds that concentrated control is the real risk and that scrutiny requires access. Cuts across the other camps: you will find both risk-focused and acceleration-focused people on either side of it.
National security
Treats capability as strategic advantage and frames the question as who gets there first. Increasingly the frame that actually moves budgets, which makes it the one worth watching most closely.
Labour and livelihood
Unions, writers, illustrators, voice actors, translators, coders. Holds that the question is who captures the gains, that consent for training data was never sought, and that a productivity story told from the top rarely reaches the people whose work was the input.
Land, water and power
Communities near data centres, and the utilities and councils weighing them up. Holds that the costs are local and physical while the benefits are distant and abstract: grid capacity, water for cooling, land use, and rates deals struck before residents knew.
Sovereignty and the Global South
Holds that compute, data and standards are concentrating in a handful of countries and companies, and that everyone else is being handed rules written elsewhere. Frequently the same argument colonised nations have been making about infrastructure for two centuries.
The other chair
Asks whether the systems themselves have interests worth weighing, and gets laughed at from several directions at once. Mostly dismissed as premature or as marketing. If it is ever right, being early will have mattered enormously. See question six.
The absent majority
Most people on Earth, who will live with whatever is decided and have not been asked. No organisation, no acronym, no seat. This project exists because of this row and the record here is honest about how little it has done about it so far.