Cryptographer Matthew Green doubts AI safety sandboxing in reply to Ezra Klein's 14 points
matthew_d_green · x · 2026-10-15
Cryptographer Matthew Green responded to Arvind Narayanan's 14 AI safety points from the Ezra Klein show. Narayanan argues intelligence isn't power — an AI system's power comes from the real-world permissions we choose to grant it, not the model itself.
Green took issue with the reliance on sandboxing, saying the point "won't age well": containing capable AI systems via sandboxing alone is unlikely to hold up. The exchange highlights a core disagreement over whether safety should focus on model internals or on external permissions and deployment environments.
Related event: AI Safety Researcher's 14 Points on Ezra Klein Podcast Spark Debate(3 posts)→
More from AGI Musings
- Funniest Gary Marcus vindication scenario: we needed AGI to invent neurosymbolic AI — akbirthko · 2026-10-15
- Scientists call for international body to oversee advanced AI — TobyWalsh · 2026-10-15
- Chollet: intelligence is rarely the bottleneck — iterated contact with reality is; Musk signals agreement — skdh · 2026-10-15
- Google details Geospatial Reasoning agent combining Earth AI models to outperform standalone LLMs — algo_diver · 2026-10-15
- Gary Marcus to speak at Bloomberg New Economic Forum, livestream available — GaryMarcus · 2026-10-15
- What does 'taste' mean in the AI era? The sum of everything you've lived — AlchainHust · 2026-10-15