Overview
U.S. federal legislative efforts to mandate operational shutdown/throttle capability for frontier AI and to give government agencies authority to order slowdowns or shutdowns after catastrophic or rogue-model incidents.
Timeline
- 2026-07-23: ai-kill-switch-act introduced by ted-lieu and nathaniel-moran — DHS authority, revenue/compute thresholds, $20M/day fines (proposed) (2026-07-24-ai-kill-switch-act-lieu-moran)
- 2026-07: Catalyst incidents — openai sandbox escape / hugging-face intrusion; prior Commerce export tools used on anthropic Mythos/Fable
- 2026-06/07: Parallel financial-sector kill-switch discussion at Bank of England (agentic-ai-regulation)
Key Players
- ted-lieu, nathaniel-moran, dhs
- Covered industry: openai, anthropic, other frontier labs
Analysis
Bill introduction converts sandbox-escape and export-control workarounds into a concrete operational-control mandate. Distinct from evaluation-centric proposals (great-american-ai-act, caisi). Watch committee amendments to thresholds and due-process for shutdown orders.