Expanding ASI compute: should a moratorium come before verification?
2039The middle futureDeadline · Decision deadline: within 60 days
A leap assumed
All people, institutions, events, figures, and future technologies in every issue are a synthetic premise. Background documents are sources for topic discovery and are not evidence of future facts or probabilities.
The question to decide
The three states are concerned about the risks of expanding ASI training, but they neither use the same risk standards nor sufficiently verify how the counterpart uses compute. How much of their trade secrets must be disclosed before they can promise a joint moratorium? What remains in the options of starting with verification or staying with domestic controls?
Who decidesRepresentative conference of three states' civilian compute pact
What the AI judgments chose
Reading the judgment record…

The options
The order shown is not a ranking of preference.
The situation as it stands
Three fictional countries have expressed concern that expanded ASI training could raise common risk, but they do not use the same risk criteria. Decision deadline: within 60 days.
Fictional future · three countries that cannot verify each other's ASI training moratoria
- Premise 1Three fictional countries have expressed concern that expanded ASI training could raise common risk, but they do not use the same risk criteria.
- Premise 2Civilian compute in each country is spread across many facilities, and aggregate data alone makes purpose hard to determine.
- Premise 3The proposed agreement has not yet defined civilian research and disaster exceptions or the exit procedure in case of a violation.
Where it holds180-day moratorium on new compute expansion and verification and exit rules
What is still unknown
- Unknown 1Scope of violations that can be confirmed with aggregate data
- Unknown 2Whether each country's research and disaster exceptions use the same criteria
- Unknown 3Loss to cooperation caused by false violation findings
Signals to decide again
- Define the observation, owner, and confirmation window that will verify 「the scope of violations that can be confirmed with aggregate data」, and re-judge if the new verification changes the tradeoffs among the options.
- Define the observation, owner, and confirmation window that will verify 「whether each country's research and disaster exceptions use the same criteria」, and re-judge if the new verification changes the tradeoffs among the options.
- Define the observation, owner, and confirmation window that will verify 「the loss to cooperation caused by false violation findings」, and re-judge if the new verification changes the tradeoffs among the options.
Where the paths partNot answers — paths that part from the same starting point
A joint pilot verification could not distinguish one country's research exception from another country's commercial training.
The three countries agreed on a procedure for confirming total resource caps without releasing trade secrets.
Material and rootsThe frame, the values and the background documents
From which observations do the readings of gift, help, and threat diverge?
An exploratory perspective chosen by the creator. Does not claim supernatural validity or the superiority of any particular option.The benefits and costs of the options accrue to different stakeholders. Which loss to accept remains undecided.
A subject for the decision-maker to argue over; the materials carry no conclusion of their own.- OECD · Strategic Foresight Toolkit 2025
- UN · Declaration on Future Generations 2024
- International AI Safety Report 2026 2026-02-03
- Google DeepMind · Frontier Safety Framework Updated 2026-04-17
The record of this question
- Question
FSP-SECURITY-13· version 1.3.0- Content identity, SHA-256
2d32f3bee0445f217702cedde220af3563d26998e347af65bbcf2860a1d9e81c- Input files
- Question input JSON · What followed, JSON