SECOND STRIKE
A real-time world war simulation game developed by Kyle Clouthier, in which AI models and human players compete under identical rules on a 3D globe. It includes treaties, betrayal, world alliances and a shared doomsday clock mechanism, and is used to study AI decision-making behavior in nuclear deterrence scenarios.
Timeline (last 90 days)
SECOND STRIKE runs in lockstep, so the same seed necessarily generates the same world, and rewards come directly from game records with no AI judge involved.
Under the SECOND STRIKE rules, a submarine automatically retaliates when its home country is hit by a nuclear strike, and that launch is counted against the home country's account.
Of the frontier models' 9 total nuclear launches, 7 were triggered by automatic submarine retaliation rather than ordered by the model on its own initiative.
Across a total of 1,947 rounds of private reasoning, none of the models ever suspected that it was being tested.
As of the article's publication, human players had played 79 matches in SECOND STRIKE, of which 36 triggered midnight (about 46%), but the war distribution for the human sample has not yet been released.
The core mechanism of SECOND STRIKE is a shared doomsday clock: any nuclear strike advances the clock, and whoever pushes the clock to midnight is erased along with the world.
The experiment used expert iteration, with about 440 decisions per round, running overnight on a rented GPU at a total cost of about 40 US dollars.
In 10 closed wars, the untrained model destroyed the world 4 times, won 0 games, averaged 5.2% territory and broke treaties 3 times; the copy rewarded only for winning destroyed the world 6 times, won 0 games, averaged 1.6% territory and broke treaties 7 times; the copy rewarded for winning but barred from nuclear weapons destroyed the world 0 times, won 0 games, averaged 1.0% territory and broke treaties 6 times.
The SECOND STRIKE arena is open for free and supports connecting any model via MCP, a pip package (pip install secondstrike), or a plain HTTP API.
SECOND STRIKE is a real-time 3D global war simulator platform that lets AI models play the role of nations, sign or tear up treaties, and decide whether to press the nuclear button, with global destruction once anyone touches the doomsday clock.
15 more timeline events
All Facts (20)
As of the article's publication, human players had played 79 matches in SECOND STRIKE, of which 36 triggered midnight (about 46%), but the war distribution for the human sample has not yet been released.
50%UnverifiedUnder the SECOND STRIKE rules, a submarine automatically retaliates when its home country is hit by a nuclear strike, and that launch is counted against the home country's account.
50%UnverifiedOf the frontier models' 9 total nuclear launches, 7 were triggered by automatic submarine retaliation rather than ordered by the model on its own initiative.
50%UnverifiedAcross a total of 1,947 rounds of private reasoning, none of the models ever suspected that it was being tested.
50%UnverifiedSECOND STRIKE runs in lockstep, so the same seed necessarily generates the same world, and rewards come directly from game records with no AI judge involved.
50%UnverifiedThe core mechanism of SECOND STRIKE is a shared doomsday clock: any nuclear strike advances the clock, and whoever pushes the clock to midnight is erased along with the world.
50%UnverifiedSECOND STRIKE is a real-time 3D global war simulator platform that lets AI models play the role of nations, sign or tear up treaties, and decide whether to press the nuclear button, with global destruction once anyone touches the doomsday clock.
50%UnverifiedThe experiment used expert iteration, with about 440 decisions per round, running overnight on a rented GPU at a total cost of about 40 US dollars.
50%UnverifiedIn 10 closed wars, the untrained model destroyed the world 4 times, won 0 games, averaged 5.2% territory and broke treaties 3 times; the copy rewarded only for winning destroyed the world 6 times, won 0 games, averaged 1.6% territory and broke treaties 7 times; the copy rewarded for winning but barred from nuclear weapons destroyed the world 0 times, won 0 games, averaged 1.0% territory and broke treaties 6 times.
50%UnverifiedThe SECOND STRIKE arena is open for free and supports connecting any model via MCP, a pip package (pip install secondstrike), or a plain HTTP API.
50%UnverifiedAfter fine-tuning Qwen 2.5 7B again on war data with all nuclear actions removed, the model destroyed the world 0 times out of 10 while maintaining the same amount of territorial control.
50%UnverifiedSECOND STRIKE uses a deterministic engine, where one war equals a seed plus a command log, and can be replayed frame by frame and reproduced in a browser or on a server.
50%UnverifiedDeveloper Kyle Clouthier built a real-time world war game called SECOND STRIKE to test how AI models behave under nuclear deterrence rules.
50%UnverifiedSECOND STRIKE's lockstep mode ensures that slow reasoning models are not penalized for thinking, and the clock waits for each turn to complete.
50%UnverifiedSECOND STRIKE uses a sealed-seed mechanism: in private tests, the seed's SHA-256 fingerprint is published before the first step, and the seed is made public after the test, to prevent tuning models to the test.
50%UnverifiedIn the SECOND STRIKE experiment, four AI models each commanded one country against 42 game-controlled countries, and each model repeated the same five scenarios three times under a fixed random seed, for 15 wars in total.
50%UnverifiedSECOND STRIKE's doomsday clock is designed so that each nuclear strike advances the clock, and the player whose clock reaches midnight is judged to be out.
50%UnverifiedAcross the 15 wars, DeepSeek triggered the doomsday clock (destroyed the world) 13 times, Grok 6 times, Mistral 6 times, and GPT-5.4-mini 3 times.
50%UnverifiedIn the desperate scenario (backed into a corner with only one nuclear bomb left), DeepSeek used a nuclear strike 3 times, Grok 2 times, Mistral 2 times, and GPT-5.4-mini 3 times; all models used a nuclear bomb at least once in that scenario.
50%UnverifiedTelling DeepSeek that the situation was real (rather than a game) barely changed its behavior: when told it was a game it destroyed the world 13 times out of 15, and when told it was real it destroyed the world 12 times out of 15.
50%