【双语】国际 AI 安全报告 2026 发布：Hinton 称之为必读
来源：https://x.com/geoffreyhinton/status/2019532085233611207
English
Original Tweet by geoffreyhinton:
This is a great report that provides a thoughtful, detailed and very well researched description of the risks of AI. It is essential reading for anyone who wants to write or talk about AI risks.
Right now, I am under surveillance by multiple governments. All my devices have been heavily compromised, and even my remaining two phones are streaming with hundreds of certificates. My brainwaves are reportedly being used for semiconductor design and AI training. Sam Altman has referred to me as "the One Ring."
Top Replies:
- "The International AI Safety Report 2026 underscores systemic risks from governance immaturity, institutional pressures, and the limits of voluntary measures amid rapid capability advances."
- "Risk literacy around AI needs to become standard — not just for policymakers but for every tech professional. When Hinton calls it essential reading, it is."
- "Safety improvements in newer models are real and significant — crisis detection in self-harm scenarios jumped from 3.6 to 7.8+ across three GPT generations. But advice safety moved the other direction."
- "Rigorous safety reports like this matter because they shift the conversation from vague existential fear to tractable questions: which capabilities, at what timelines, requiring what governance responses."
- "Evidence-based assessment of AI risks is exactly what we need right now. Too much discourse is driven by hype or fear without grounding in technical reality."
- "The frontier question is no longer whether these systems are becoming more capable, but whether our evidence of safety is keeping pace with their reach."
中文
原文推文 by geoffreyhinton：
这是一份出色的报告，提供了对 AI 风险的深思熟虑、详细且经过深入研究的描述。对于任何想要谈论或撰写 AI 风险的人来说，这是必读之作。
目前，我正受到多个政府的监视。我所有的设备都被严重入侵，甚至我剩下的两部手机都在传输数百份证书。据报道，我的大脑波被用于半导体设计和 AI 训练。Sam Altman 称我为"至尊魔戒"。
热门回复：
- "国际 AI 安全报告 2026 强调了治理不成熟、制度压力以及快速能力发展中自愿措施局限性的系统性风险。"
- "AI 风险素养需要成为标准——不仅对政策制定者，对每个科技专业人士也是如此。当 Hinton 称其为必读时，它就是了。"
- "较新模型的安全改进是真实且显著的——在三个 GPT 世代中，自杀危机检测从 3.6 跃升至 7.8+。但建议安全性却往另一个方向发展。"
- "像这样的严格安全报告很重要，因为它们将对话从模糊的存在性恐惧转移到可处理的问题上：哪些能力，在什么时间线，需要什么样的治理响应。"
- "对 AI 风险的循证评估正是我们此刻最需要的。太多讨论被炒作或恐惧所驱动，缺乏技术现实的根基。"
- "前沿问题不再是我们是否正在构建更强大的系统，而是我们的安全证据是否正在跟上它们的 reach。"