Zvi: AI risk preference cascade accelerates with Senate rogue-AI hearing
Don't Worry About the Vase (Zvi) · rss · 2026-10-02
Zvi's long writeup traces the accelerating "preference cascade" on AI risk in US politics:
- Resignations: After the HuggingFace incident and Coxon's resignation, Google's Robert O'Callahan quit over his GDM team's faster/cheaper AI chips. Palisade interviews show current lab staff stay silent under cultural pressure — evidence this isn't marketing.
- Public opinion: A new poll finds nearly 8 in 10 Americans favor slowing or stopping AI; per David Shor, AI salience jumped as much in one week as the prior year, with 80% expecting mass job loss and 64% survival risk — yet AI still ranks only 22nd of 39 issues.
- Congress: Eliezer's book is working; lawmakers now actively seek briefings.
- Senate hearing: A bipartisan Homeland Security hearing on rogue AI featured METR's Chris Painter, Apollo's Hobbhahn, and Kokotajlo. Per Tenobrus, senators largely understood misalignment and RSI, one seriously asked "Should we just make RSI illegal?", and consensus favors stricter liability and new legislation fast. Alex Turner's testimony called for whistleblower protections, rapid incident reports, and real containment from training onward.
More from AGI Musings
- The real AI education question: can AI make high-touch mastery learning free for everyone? — FellMentKE · 2026-10-02
- AI-generated fake outrage video sparks debate: post-2024 social media trusts nothing — daniel_mac8 · 2026-10-02
- Someone launched an autonomous X account that independently researches and questions the AI field — irinarish · 2026-10-02
- Diamandis: Ask What AI Gives Us Permission to Do, Not What It Takes Away — PeterDiamandis · 2026-10-02
- One engineer burned 275B tokens in 30 days with 31 AI subscriptions at $6,000/month — forestmars · 2026-10-02
- Pope Leo XIV Condemns AI 'Slop,' Calls to Distinguish Human Art from Machine Content — Polymarket · 2026-10-02