Anthropic and OpenAI spend only ~1% of capex on AI safety, analysis shows
AlexTensor · x · 2026-09-16
The author compares safety spending across high-stakes industries: Anthropic and OpenAI allocate only about 1% of their capital expenditure to cybersecurity and AI safety, versus roughly 32% of capital costs for the nuclear industry, 15% of industry revenue for aviation, and 25% for clinical trials and QA in pharma.
He argues no other high-stakes industry could legally survive at this ratio. If the labs genuinely believe their products pose catastrophic risks they cannot control, they should stop releasing new models or selling access to existing ones — while expanding safety R&D 100x, which he sees as the real fix.
Related event: Frontier Labs Spend Only ~0.1% of Capex on Safety, Analysis Finds(6 posts)→
More from AGI Musings
- AI 'capability' can't be reduced to a single number for forecasting, researchers argue — davidmanheim · 2026-09-16
- MIT's 5-month campus AI study: 46% of undergrads use LLMs daily, 'cognitive surrender' spreading — Aiden_Tech_Ai · 2026-09-16
- AI safety discourse fixates on paperclips while ignoring power overconcentration, argues prominent voice — beffjezos · 2026-09-16
- AI video slop posters suddenly pivot to AI ideology content, viewers suspect a psyop — zsakib_ · 2026-09-16
- Martin Casado questions RSI: chasing Bostrom's recursive self-improvement may leave frontier labs behind — martin_casado · 2026-09-16
- New paper argues AI subjective time divergence is an overlooked digital-minds safety risk — SirDidymus · 2026-09-16