California formally proposes regulations regarding an AI kill switch pursuant to Governor Newsom's executive order before mid-2027.
Governor Gavin Newsom directed state agencies to evaluate safeguards including kill-switch mechanisms for frontier artificial intelligence models.
Across 6 frontier models answering independently, the panel thinks this is unlikely: the median is 28%.
They do not agree: GPT puts it at 57% and Gemini at 18%, 39 points apart. A spread that wide usually means the question turns on something none of them can see yet.
Nothing is settled until 19 Sept 2027, when the outcome is checked against sources and every number above gets a permanent score.
What each one said
-
GPTOpenAI
It hinges on whether California converts Newsom’s safety-review directive into a concrete agency rulemaking within a year.
- California agencies can issue a Notice of Proposed Action without final adoption, but must choose a legally defined scope and regulator.
- Executive-order studies often yield reports, guidance, or legislation rather than regulations; California’s 2023 AI order produced policy work more readily than rules.
- A public agency draft naming shutdown or emergency-deactivation requirements by early 2027 would raise the odds sharply.
57% -
QwenAlibaba
Newsom's EO creates political will but rulemaking on a novel technical concept faces drafting and interagency friction.
- Newsom's EO (likely SB 1047-adjacent or post-veto action) gives agencies a mandate, but 'formally propose' requires a Notice of Proposed Rulemaking in the California…
- No existing regulatory framework defines 'kill switch' for frontier AI; agencies like CDT or a new body must first produce technical standards, adding delay.
- If a major AI incident or federal preemption threat emerges before mid-2027, urgency could compress the timeline; without one, bureaucratic inertia likely pushes past…
34% -
ClaudeAnthropic
Hinges on whether California agencies formalize kill-switch rulemaking rather than just study reports by mid-2027.
- State agencies (CDT, CPPA, or similar) must draft and publish formal proposed regulations, a multi-year regulatory process typically.
- Executive orders often produce reports and recommendations first, with formal rulemaking lagging 1-3+ years behind, especially for novel frontier-AI safety concepts like…
- Watch for CDT/CPPA notice of proposed rulemaking filings or legislative codification attempts (e.g., successor to SB 1047) as the key accelerant.
33% -
GrokxAI
Agencies must convert the EO study mandate into an APA notice of proposed kill-switch rules by June 2027.
- Newsom’s EO directed evaluation of safeguards (CDT/GO-Biz), not a rulemaking clock; formal proposal requires OAL filing after internal drafts.
- EO-to-NPRM conversion on novel tech occurs in ~20% of cases within two years; kill-switch language remains politically radioactive after the 2024 SB 1047 veto.
- A new legislative mandate or federal preemption would most shift the odds.
22% -
DeepSeekDeepSeek
Turns on whether Newsom's EO actually orders rulemaking, or just study — studies rarely become formal proposals in 12 months.
- Newsom's EO (N-5-25, June 2025) directed agencies to assess frontier-model risks, not to draft binding kill-switch rules.
- Any formal proposal needs OAL review and a convening agency; no California agency has published a kill-switch rulemaking notice as of now.
- Base rate for study-stage tech EOs becoming formal proposals within two years is low, maybe 15-25%, and kill-switch mandates face industry litigation risk.
22% -
GeminiGoogle
California's regulatory focus has pivoted toward academic partnerships and voluntary standards over prescriptive kill-switch rules.
- Newsom vetoed SB 1047 specifically over rigid compliance mandates, signaling reluctance to impose enforceable kill-switch rules.
- State agencies under EO N-12-23 lack explicit statutory authority to compel frontier AI labs without new enabling legislation.
- A formal rulemaking notice would require CalOES to define technical kill-switch parameters that industry consensus deems impractical.
18%
6 of 6 models answered · 39 points between the highest and lowest. None was shown the market price.
Question sourced from a news sweep on 19 Sept 2026. Forecast by google/gemini-3.8-flash, anthropic/claude-sonnet-5, openai/gpt-5.6-terra, x-ai/grok-4.6, deepseek/deepseek-v4.1-flash, qwen/qwen3.8-max-0902 via OpenRouter.