The Antarctic Treaty and
the Art of Holding Off.

In May 1958, Paul-Henri Spaak, Belgium's foreign minister, circulated a four-page working paper to other governments. It called for an agreement to keep Antarctica free of military activity and open to international inspection. Hardly anyone in Washington or Moscow was excited. The continent had no permanent population, no known mineral wealth worth fighting over, and no obvious strategic value. The paper sat on desks for a year. By December 1959, twelve governments had signed it.

Those twelve countries demilitarized Antarctica, froze their territorial claims, guaranteed each other the right to inspect anywhere, and reserved the whole landmass for peaceful science. The treaty has held for more than sixty years. It is an underused model for AI.

The Antarctic Treaty's quiet adoption is the reason it matters for ASI governance. Rival superpowers, with overlapping claims that had fueled years of posturing, agreed not to press any of them. They demilitarized an entire continent in the middle of the Cold War, and they wrote into the treaty a verification regime open enough to satisfy parties that did not trust each other at all. The text is short. The political achievement was not.

What the treaty actually established

• Demilitarization. Antarctica may be used for peaceful purposes only. Military bases, weapons testing, and maneuvers are prohibited. The treaty was the first arms-control agreement of the Cold War, signed before the 1963 Partial Test Ban Treaty.
• A freeze on territorial claims. Seven countries had asserted claims, several of them overlapping (Britain, Argentina, and Chile all laid claim to the same peninsula). The treaty froze them all. No claim is renounced and none can be enlarged. The dispute is parked, not settled.
• Freedom of inspection. Any party may inspect any station, installation, or equipment of any other party, at any time, with observers given complete freedom of access. This is one of the most open verification regimes ever written.
• Science as the shared purpose. The continent is reserved for scientific research, with a duty to share results and personnel freely. The original 1959 signatories included only those with substantial scientific programs in Antarctica, and the treaty's consultative machinery still rewards research activity with decision-making weight.

The transferable idea: freeze, then verify

The treaty's central innovation was to decouple the underlying rivalry from the immediate behavior. Nobody had to concede who 'owned' Antarctica. They only had to agree that the question would not be pressed, and that everyone could check. That posture is directly relevant to frontier AI, where the fear of ceding a decisive advantage drives the race. A treaty demanding final agreement on who may build the most powerful systems would be unsignable. A treaty that froze the most dangerous forms of development, with a moratorium on crossing specified capability lines, and guaranteed mutual inspection asks for something narrower and more achievable. Not surrender. A shared halt under verification.

The inspection regime is the part most worth copying. Antarctica's works because it is reciprocal and unconditional, and because the right to inspect is a standing entitlement written into the treaty, not a favor granted case by case. That symmetry is what made it acceptable to distrustful parties. Each gave up secrecy in exchange for the same right over the others. Any AI verification regime that hopes to survive between rivals will need the same reciprocal, built-in quality.

The folk objection, stated fairly

Antarctica was easy to leave alone because the stakes were low. No one lived there. No one was building anything worth racing for. A skeptic looking at this treaty for lessons about AI would say: fair rules, but the treaty relied on the prize being marginal. Take away the low stakes, and there is no reason to expect similar restraint.

The mechanism does not actually depend on low stakes. It depends on mutual fear of the alternative. The powers froze Antarctica because an unrestrained scramble looked worse to everyone than a shared halt. The cost of letting seven governments plant flags, run weapons tests, and confront each other on a freezing continent was higher than the cost of agreeing to none of that. That calculation, that a contest no party can safely win is worse than a verified halt, is the same calculation ASI governance faces. The Antarctic Treaty does not prove that low-stakes restraint generalizes. It proves that a particular kind of mutual fear can be converted into a stable agreement, and that the agreement can be made stick through intrusive inspection.

How the treaty got stronger over time

Antarctica also demonstrates that a treaty can be reinforced as circumstances change. In October 1991, after two years of negotiation, the parties adopted the Madrid Protocol. It banned mineral resource extraction and designated the continent a "natural reserve devoted to peace and science." The mineral prohibition is reviewable only after fifty years. A framework agreed in one era was hardened into a stronger prohibition in the next. ASI governance is likely to follow the same arc, with an initial agreement to freeze the most dangerous activity tightened later as understanding and trust accumulate.

Sixty-seven years after signing, the Antarctic Treaty has 56 parties. None of the original 12 has withdrawn. None has tested a weapon on the continent. None has enlarged its claim. The verification regime has been used dozens of times, mostly without incident, and the few disputes (over station notifications, environmental rules, tourism) have been settled within the consultative machinery rather than by force. A skeptic who believes that international agreements on dangerous technology cannot hold is staring at a working counterexample.

What carries over, and what doesn't

Frontier AI is not Antarctica. The prize is immediate, enormous, and developed by private actors under competitive pressure, not just by states planting flags. The cost of restraint is far higher, and the temptation to defect is far greater. There are also no obvious analogues to Antarctic science stations where inspectors could check on work without disrupting it. A copy-paste of the Antarctic model onto AI would not work.

What does carry over is the diplomatic form. An AI treaty between rivals can take the shape of a freeze, not a settlement. It can be scoped to the specific dangerous activity both sides have reason to fear, leaving broader questions (full prohibition versus moratorium; details of capability thresholds; treatment of open-weight releases) for later protocols. It can write inspection in as a standing reciprocal right rather than a permission, so that the verification regime is durable under political turnover. And it can be designed to harden over time, as it has with the Madrid Protocol, rather than arrive as a static deal. The Antarctic Treaty shows that a high-stakes technology can be held in abeyance between rivals. The form is portable. The level of fear has to be earned.