# 评Anthropic员工AI灭绝人类论

- 来源：Chubby♨️ (@kimmonismus)
- 发布时间：2026-09-09 19:33
- AIHOT 分数：48
- AIHOT 链接：https://aihot.news/items/cmtu11den0ukarofpdufo31b5
- 原文链接：https://x.com/kimmonismus/status/2097649566677885089

## AI 摘要

Kim发文回应Anthropic员工认为AI有超10%概率灭绝人类的观点，称其缺乏依据。她指出AI竞赛如同囚徒困境，各国难以放缓，但AI本身无毁灭意图，风险在于应用方式而非技术本身。她呼吁停止恐惧宣传，进行事实性讨论。

## 正文

We need to talk about this. Anthropic employees believe there is a realistic chance that AI will wipe out humanity, greater than 10%.

"Jacob is correct here—we really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade."

The situation has reached a point where employees are even leaving these corporations because, in their view, none of the "frontier labs" are acting responsibly.

I would like to share my thoughts on this. Let me start with a metaphor I have used often: Pandora’s box has been opened. There is no turning back.

What do I mean by this? Within the current systemic landscape, there is a fierce competition for future dominance. Undoubtedly, the USA is currently the world’s strongest nation, with China as its greatest rival. AI is regarded as a strategic technology that impacts national sovereignty and is, therefore, subject to political control. Because this technology is treated as *the* technology of the future - and because it is assumed that the nation possessing the best AI (including RSI, etc.) will emerge as the global hegemon - there is no possibility of any party slowing down. It is essentially a prisoner's dilemma.

The only possibility would be for all parties to view the technology and the race toward the singularity as so threatening that they agree to slow down collectively and establish joint accords, similar to nuclear arms treaties. However, there is currently no sign of that happening.

Regardless of that, the question remains: how exactly is the extinction of humanity being justified? What are the arguments for it? Certainly, we have seen what models are already capable of today. Yet, it is difficult to posit a genuine intent to cause destruction. Such destruction would presumably result from one party deploying the technology against another, much like the use of a nuclear weapon. In that case, however, the root cause of the danger would not be the technology itself, but the underlying systemic race.

In this respect, we need a sober assessment of the reasons that might actually support such a scenario. The idea that an AI might go "rogue" - simply because it feels like it, views the human species as subservient, and *wants* to wipe us out - requires a solid justification. What kind of will would underlie this? Why an intention to destroy rather than to collaborate? Can an artificial (!) intelligence even possess a will? To the best of my knowledge, these questions remain unresolved. Consequently, I refrain from propagating the idea of a major threat. The only danger - as I have previously stated - lies in how the technology is applied, much like any technology can be put to military use. However, this is not an inherent characteristic of AI itself.

In this regard, I would urge us to step back from fear-mongering. AI has the potential to improve all our lives - permanently. There is hardly any doubt about that anymore. It can, of course, be misused for nefarious purposes; that, too, is clear. Yet, inferring the extinction of humanity from this - and even assigning probabilities to such an outcome - strikes me as far-fetched. I would very much welcome a factual discussion on the matter and would be happy to explore all the arguments in detail.

### 引用推文

> Evan Hubinger：Jacob is correct here—we really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade. I believe Anthropic is tryi...