Skip to content
LAB IN FORMATION / a gathering point for AGI readiness & AI safety

the threshold is the work.

a lab is forming for the people willing to work on what safe superintelligence would require — especially where the necessary conditions feel impossible to achieve.

necessary does not become optional because it is difficult.

alignment that survives a change in incentives. evidence that survives independent scrutiny. institutions that can stop a deployment when stopping becomes expensive. coordination that holds after agreement ends. these are the conditions we want to turn into shared research.

ASI Readiness is forming to connect the people working on models, minds, institutions, values & global coordination through concrete problems, reproducible experiments and decisions the evidence can actually change.

the necessary thresholds

PROPOSED RESEARCH AGENDA
01 / evidencewhat would justify the claim — and what would force us to withdraw it?02 / controlwhat still works when the system understands the safeguard?03 / institutionswho can act on the evidence when acting becomes costly?04 / coordinationwhat holds when incentives diverge and trust is incomplete?05 / human agencycan the people affected meaningfully contest the decision?

arguments at the threshold

all six essays →
01founding thesis / 4 min read

the threshold is the work

A founding argument for a lab willing to work on the conditions safe superintelligence would require, especially where those conditions seem institutionally impossible.

02evidence & evaluation / 4 min read

a benchmark is not a safety case

The missing argument between passing an evaluation and deserving authority in the world.

03scientific integrity / 4 min read

compute is not a theory of truth

A scientific claim owes us an exact statement, inspectable assumptions, reproducible evidence and visible human credit, whatever produced it.

04alignment & control / 4 min read

alignment has to survive knowing it is being tested

The gap between observed compliance and robust control is a research program, not a detail to hide in the appendix.

05institutional design / 4 min read

the right to stop must survive the cost of stopping

A proposed institutional test for safety authority: can the evidence change the decision when doing so becomes expensive?

06coordination & legitimacy / 4 min read

coordination that only works when everyone agrees is not coordination

Readiness needs agreements that survive unequal incentives, contested evidence and the temptation to defect.