Safety researchers propose $10M compute pre-release tests for destabilizing AI capabilities
Jsevillamol · x · 2026-09-20
Greg Burnham proposes frontier labs spend $10M of compute in pre-release testing on easily verifiable "destabilizing" innovations like polynomial factorization. Jsevillamol agrees, noting inference scaling favors defense only if compute holders actually commit a significant budget to it, and argues testing shouldn't stop at verifiable domains: try overcoming governments, designing lethal pathogens, and other hard-to-verify risks.
More from Safety
- OpenAI subpoenaed researcher's communications over California AI safety law SB 53 — trevposts · 2026-09-21
- Flock Safety offers employee buyouts amid backlash over surveillance tech — Polymarket · 2026-09-21
- Emad Mostaque: frontier weights are becoming national security assets, expect ITAR-style export rules — ccerrato147 · 2026-09-21
- Report: Anthropic denied UK AI Security Institute pre-release access to 'Mythos 5.1', a first — ccerrato147 · 2026-09-21
- Fake YouTube AI trading bot tutorial drains 224 wallets of $517,000 in ETH — 4KTV · 2026-09-21
- Satire imagines an AI license regime: $20k fines and 10-year jail for unlicensed use — TinfoilTricorn · 2026-09-21