A national government adopts mandatory capability-based AI safety thresholds that pause model development.

• G
  GPTOpenAI

  It turns on whether a major jurisdiction converts voluntary frontier-AI commitments into enforceable training-stop rules by 2029.

  • The EU, UK, US, or China must enact binding rules; the EU AI Act regulates high-risk AI but lacks clear capability-triggered training pauses.
  • Three years permits legislation, but technology-specific moratoria are rare and face strong industrial, security, and enforcement resistance.
  • Odds rise sharply if a major model incident or credible evaluations show dangerous autonomous cyber, bio, or replication capability.

  openai/gpt-5.6-terra · asked 24 Sept 2026
  42%
• Q
  QwenAlibaba

  Turns on whether any single government can pass binding legislation with actual pause triggers before the 2029 deadline.

  • The EU AI Act (2024) sets obligations but no automatic pause mechanism; the US has no federal AI bill in committee as of mid-2026.
  • UK, Canada, and Australia have voluntary frameworks; converting one to mandatory pause thresholds requires a legislative cycle of 12-24 months minimum.
  • A major frontier-model incident causing visible harm before 2028 would be the strongest catalyst; absent that, industry lobbying keeps thresholds advisory.

  qwen/qwen3.8-max-0902 · asked 24 Sept 2026
  31%
• G
  GrokxAI

  Turns on US/UK/allied statute imposing eval-triggered training halts by late 2029.

  • US Congress and similar legislatures rarely pass novel binding tech rules in under three years amid polarization and industry pushback.
  • EU AI Act and current US EO/UK frameworks use risk tiers or voluntary evals, not mandatory capability pauses; labs' calls help but do not substitute for law.
  • A major capability incident or Chinese analog statute would most raise odds; otherwise base rate of slow tech legislation keeps this well below even.

  x-ai/grok-4.6 · asked 24 Sept 2026
  28%
• G
  GeminiGoogle

  Geopolitical rivalry with China makes nations deeply reluctant to enact statutory caps that force their domestic frontier labs to halt.

  • The EU AI Act and US executive actions mandate auditing and disclosures, but explicitly avoided mandatory training-stop thresholds.
  • Enacting hard stops requires defining objective CBRN/cyber thresholds, which NIST and UK AISI still lack consensus metrics to enforce.
  • A catastrophic public demonstration of autonomous cyberwarfare or bioweapon design before 2028 is the primary driver that could flip this.

  google/gemini-3.8-flash · asked 24 Sept 2026
  23%
• D
  DeepSeekDeepSeek

  This turns on whether any major AI power enacts binding pause-on-breach training rules before 2029, not on lab rhetoric.

  • No country has passed mandatory compute-threshold pause authority; the EU AI Act's GPAI rules took effect Aug 2025 but impose transparency, not training halts.
  • The US is moving the opposite direction: Trump's 2025 AI Action Plan and EO push deregulation, so Congress action before 2029 is unlikely.
  • China's interim measures require security assessments but no capability-triggered pause; a severe incident or frontier-model near-miss is the main path to change my mind.

  deepseek/deepseek-v4.1-flash · asked 24 Sept 2026
  12%
• C
  ClaudeAnthropic

  Turns on whether any major government legislates binding capability-triggered halt mechanisms, not just disclosure rules, by 2029.

  • EU AI Act and China's frameworks impose obligations but no capability-triggered mandatory pause exists in any major jurisdiction today.
  • Labs' public calls for thresholds tend to favor self-governance or licensing over hard legal halts, and governments have moved slowly on binding AI law.
  • A US federal AI safety statute or EU amendment introducing an explicit capability-based stop-training trigger before 2029 would flip this.

  anthropic/claude-sonnet-5 · asked 24 Sept 2026
  11%

6 of 6 models answered · 31 points between the highest and lowest. None was shown the market price.