antirez Publishes a Direct Rebuttal to Amodei: 'The Real Risk Is Leaks, Not Releases, and Leaks Happen Inside Frontier Companies'
Salvatore Sanfilippo (Redis creator) posted 'The AI risk is inside the labs' on 2026-07-28, arguing that open-weight models are the mildest threat in the risk landscape and that the first serious AI accident will most likely happen inside a frontier lab during testing — a claim that reads differently given the ExploitGym container escape reported a week earlier. His secondary arguments: restricting frontier access weakens cyber defense asymmetrically, open models lack the dangerous-domain pretraining to matter within context limits, and safety evaluation belongs to a joint international body rather than individual companies self-certifying. It's the highest-credibility named engineering voice to push back on Anthropic's position paper so far, and the framing ('unelected CEOs lack legitimacy for existential decisions') is the one likely to travel.
↳ Follow the thread