Vinod Khosla: Open Source Models Too Costly and Dangerous, Urges US-China Treaty to Constrain Them
rohanpaul_ai · x · 2026-11-05
Vinod Khosla, founder of Khosla Ventures, shared striking views on open source models in a new WSJ interview:
- Open source is far costlier and more dangerous than closed source, he argues.
- He has recommended a US-China treaty on open source to David Sacks and the White House: constraining open models from doing wild things (like designing another bad virus) is in both nations' mutual interest — and not just a nation-state issue, since an individual at home could design and release a bad virus.
- Closed source has oversight; open source doesn't: OpenAI, Anthropic and Google know what people use their models for, but open models can run anywhere, so safety harnesses can't be kept in place.
- His proposal: restrict open source to certified clouds where certain bad uses are blocked — workable in the US, China, and everywhere else.
Full video on the WSJ Events YouTube channel.
More from AGI Musings
- Prediction market roundtable: CLOBs let market makers profit, AMMs lose them money — devanshmehta · 2026-11-05
- India's tech firms optimized for cheap labor for decades — now AI makes them pay the cost — arpit_bhayani · 2026-11-05
- AI safety researchers clash over whether instrumental convergence predicts agentic risk — davidmanheim · 2026-11-05
- Caught Between AI High Priests and the Luddite Left: One Poster's Proud Centrism — teodorio · 2026-11-05
- The Next Bottleneck Isn't Compute — It's Human Attention — manosaie · 2026-11-05
- LessWrong's flaw: treating 'being less wrong' as an identity, says blogger — StewartalsopIII · 2026-11-05