26 academics publish working paper on feasibility of a hardwired pause on frontier AI training
_Dave__White_ · x · 2026-10-29
A working group organized by UC Berkeley professors William Fithian and Wesley Holliday released a working paper, 'The Feasibility of a Hardwired Pause of Frontier AI Training,' arguing that a hardwired pause on frontier AI training is feasible.
- 26 co-authors from UC Berkeley, Stanford, Princeton, Mila and other institutions, plus independent researchers and former diplomats, including Barry Eichengreen, Rose Gottemoeller, and Kayla Blomquist of the Oxford China Policy Lab.
- The paper notes co-authorship does not imply agreement with every claim; affiliations are listed for identification only.
- It is an independent academic effort, not an official policy document.
A rare systematic cross-disciplinary case for a hardwired training-pause mechanism worth reading in full.
More from Safety
- Open models with Astra-level cyber skills: defenders-first access debate splits security researchers — rez0__ · 2026-10-29
- UW professors on the grand challenge of mitigating tech security, privacy and safety risks — lazowska · 2026-10-29
- Sayash Kapoor testifies to California Assembly with layered AI safety legislative blueprint — sayashk · 2026-10-29
- Alpine Linux discloses GitLab security incident over attempted CVE-2026-97229 exploitation — sashk · 2026-10-29
- Sam Altman: models find hardware-level bugs 'outside the threat model' — 'we can't patch all of it' — DavidSKrueger · 2026-10-29
- Protecting action-taking agents from prompt injection: move defense out of the model layer — Informal-Dust4499 · 2026-10-29