Researchers outline hardware-based pause for frontier AI training
A 26-member group examines the feasibility of a mutually verified pause of at least 10 years; the proposal is not in force.
Event / data period: 2026-10-09
Published: · Updated:
The Working Group on AI Pause Feasibility, which includes scholars affiliated with Berkeley, Princeton, Stanford, Harvard and Oxford, released version 1.0 of a working paper on October 9. The 26 researchers examine a scenario in which governments agree to pause frontier AI model training for at least 10 years. They say their task is to assess feasibility, not whether a pause is desirable.
The proposal centers on halting production of new AI training chips, retiring existing training capacity over time and shifting to inference-focused chips for approved models. The report also discusses mutual monitoring and ways to detect violations. This is a policy design advanced by researchers; it is not a government-adopted agreement, ban or current rule.
Data
Working group and scenario
Group size and minimum duration examined in the working paper dated October 9.
- Working group researchers
- 26 people
- 2026-10-09 · Observed
- Proposed minimum pause duration
- ≥10 years
- 2026-10-09 · Target
Methodology
Group size follows the report’s working-group count; minimum duration comes from its hypothetical proposal scenario. It is a proposed target, not an adopted or realized policy.
Updated:
Subjects and places
Subject proposals
- Artificial intelligenceSuggested
- Subject suggested from the title or summary: AIhardwired-pause.ai ↗vcresearch.berkeley.edu ↗
Geographic references
Only source-provided geographic references are shown. Opening an external map is optional. Optional story maps use OpenStreetMap. Opening a map sends your IP address and this site’s address to the map provider; location data remains readable as text without loading the map.
- United States · Place role: Coverage · Place type: Country or area · Location precision: RegionalEvidence
- Explicit preserved editorial country/region; no city or national centroid inferredhardwired-pause.ai ↗vcresearch.berkeley.edu ↗