rationale:

The plurality probability on 'None of these before 2032-01-01' (50%) stems from the severe strictness of all three event triggers, contrasting with the heavily hedged and evolutionary reality of AI R&D. Even if autonomous agent capabilities scale to massive expenditure horizons by 2029 , labs face structural, liability, and safety disincentives against making absolute triumphal claims that eliminate human accountability. There is a substantial chance that AI models will generalize to run full, bounded experimental loops by mid-2028 , but developers have strong reasons to frame this as human-in-the-loop infrastructure. Internal R&D risks also currently command only roughly 14% of frontier-lab safety research effort , reflecting an industry not yet in a recursive emergency posture. As of August 2026, labs actively deny threshold crossings, and governments are producing reporting regimes rather than binding internal use restrictions 2 sources.

A public claim of full AI research coding automation is the most plausible first mover among the positive options (30%). Trendlines here are steep: Anthropic already reports that a large majority (over 80%) of merged production code is authored by Claude anthropic.com. However, the resolution bar requires an official statement that research coding 'requires no human coders.' Labs consistently hedge their language and market human oversight as a safety feature anthropic.com, making a categorical claim a very high hurdle.

Prospects for a binding government restriction stand at 15%. Elite demand for pacing mechanisms is visibly rising, evidenced by the 1,376 signatories to the July 2026 'Pacing the Frontier' letter blog.aifutures.org. Nonetheless, no jurisdiction currently restricts the use of AI to conduct AI research; existing frameworks focus on models as released products 44 sources. Evaluating the structural hurdles to legal interventions on internal R&D reveals that enacting a targeted, binding restriction on internal research use would require a substantial policy pivot, likely only triggered by emergency executive action after a major capability leap.

An official claim that an AI lab no longer requires human researchers at any stage is the least likely first event (5%). Not only would this naturally follow a coding-automation claim, but it also collides forcefully with the developers' own safety frameworks. Under OpenAI's Preparedness Framework and Anthropic's RSP, crossing full-substitution thresholds triggers sweeping safeguard mandates or mandatory development halts 3 sources—a massive disincentive for absolute phrasing. Furthermore, human functions like open-ended horizon planning remain persistent bottlenecks 2 sources.