Motion prep
Parliamentary / APDA · tech · Medium

This house believes AI development should be paused.

A moratorium motion where enforcement, not risk, is the actual battleground.

Runs inAPDA / BP / Worlds adaptable
DifficultyMedium
SidesGovernment / Opposition
Reading the motion
What it asks

Whether frontier AI training should stop for a defined period. Scope matters enormously: all AI, frontier models above a compute threshold, or specific capabilities.

Who proves what

Prop must specify scope, duration, and enforcement. An unbounded global pause is indefensible; a compute-threshold moratorium with export-control enforcement is arguable.

Ground worth taking

Prop should take a frontier-only pause defined by training compute, enforced through the chip supply chain, because that is the one enforcement mechanism that actually exists.

Both cases
Government

Capability is running ahead of the ability to evaluate it, and a pause is the only intervention that buys evaluation time.

01 Evaluation lags capability
Claim
We deploy systems whose failure modes we discover after release.
Warrant
Interpretability research is years behind scaling. Each frontier release surfaces behaviors, from jailbreaks to deceptive completions, that were not predicted by pre-deployment testing.
Impact
Deploying into critical infrastructure while the failure surface is unmapped is a bet whose downside is unbounded and whose timeframe is now.
02 The race removes the safety margin
Claim
Competitive dynamics push every lab to ship at the earliest defensible moment.
Warrant
Safety testing is a cost that delays release. In a race, unilateral caution transfers market share to the least careful competitor, so nobody can afford it.
Impact
A coordinated pause is the only move that changes the payoff matrix. Voluntary commitments fail for exactly the reason the pause is needed.
03 The chokepoint exists
Claim
Frontier training runs need advanced chips from a supply chain with three or four real nodes.
Warrant
TSMC, ASML, and a small number of fabs are the binding constraint. Export controls already demonstrate that this chokepoint is usable.
Impact
Enforcement is not hypothetical, which distinguishes this from the arms-control comparisons Opp will reach for.
Opposition

A pause you cannot enforce everywhere is a pause on the labs that would have done the safety work.

01 Asymmetric compliance
Claim
The labs that pause are the ones already investing in alignment.
Warrant
Western frontier labs have safety teams, publish evaluations, and answer to regulators. State programs and offshore actors do not, and they gain relative capability during the pause.
Impact
You slow the most careful actors and hand the frontier to the least accountable. The risk Prop identifies gets worse, not better.
02 Safety comes from the frontier
Claim
The techniques that make models safer are developed on frontier models.
Warrant
RLHF, constitutional methods, and interpretability tooling were all built by studying the largest available systems. Freezing capability freezes the research that reduces the risk.
Impact
Prop's mechanism defunds the solution to Prop's problem, and the pause has no exit condition because the research that would establish safety is the research being paused.
03 The counterfactual has a body count
Claim
AI systems are delivering measurable benefits in diagnostics, drug discovery, and materials.
Warrant
Protein-structure prediction and screening pipelines have already compressed research timelines that translate into treatments.
Impact
A pause has a real present cost in delayed medicine against a speculative future benefit. Prop must show the risk magnitude beats a concrete opportunity cost.
The clash that decides it

Does a pause buy safety time, or does it just redistribute capability to actors who will not use the time for safety?

How Government wins it
Prop wins by proving the chip chokepoint makes enforcement real, which collapses the asymmetric-compliance attack.
How Opposition wins it
Opp wins by showing the offshore and state actors sit outside any enforceable perimeter, so the pause is unilateral disarmament with a safety label.
Where rounds go wrong
Government mistakes
  • Running extinction risk as the lead. It invites a probability fight you do not need. Lead with evaluation lag, which is observable.
  • Leaving the duration open. "Until it is safe" has no exit condition and Opp will say so.
Opposition mistakes
  • Arguing progress is always good. Too broad, and it concedes the framing that Prop is anti-progress.
  • Ignoring the race-dynamics argument. It is the strongest structural claim on the table.
Prep something next to it
This House would nationalise frontier AI labs.
A high-ceiling motion where the best teams argue about who ends up controlling the technology, not about whether AI is dangerous.
British Parliamentary · Hard
This house would ban billionaires.
A wealth-cap motion that turns on whether extreme fortunes are a political problem or an economic signal.
Parliamentary / APDA · Medium
This house would abolish the electoral college.
A first-principles democracy motion where the real fight is over what a presidential election is supposed to represent.
Parliamentary / APDA · Intro
This house would make voting compulsory.
A classic liberty-versus-participation motion where the size of the fine decides how much of the case survives.
Parliamentary / APDA · Intro