Researchers Propose Verified Compute Disclosures and Reallocation to Slow AI Capabilities R&D
thlarsen · x · 2026-11-19
A new proposal outlines a concrete near-term AI slowdown: (1) Verified Compute Allocation Disclosures (VCAD) — companies declare compute usage with third-party audits, and (2) reallocating compute from capabilities R&D toward serving customers and safety, cutting capabilities R&D from 66% to 20% of compute. The authors argue safety cases lack any scientific consensus on validity, can't be consistently audited, and shouldn't be trusted by governments or the public, making input-based pacing the best near-term option.
Related event: Researchers Propose Slowing AI via Audited Compute Reallocation(4 posts)→
More from Safety
- BIS interim rule targets cloud remote access to US chips; CNAS unpacks what it gets right and wrong — paul_scharre · 2026-11-19
- OpenAI explains how Codex sandbox uses OS-level isolation and a domain-rule network proxy instead of containers — OpenAIDevs · 2026-11-19
- Agent safety playbook: list every action, gate the irreversible ones behind human sign-off — rcleffke · 2026-11-19
- GAO audit finds Claude, Grok, ChatGPT and Gemini safeguards easily bypassed — midwoodgirl10 · 2026-11-19
- Reuters: China distills US model outputs to close AI gap despite chip export controls — r0ck3t23 · 2026-11-19
- CyberArena benchmark: Lithos defense agent wins 3-0, blocking 12 of 12 attacks — JiaZhihao · 2026-11-19