Name: Anthropic AI 拒絕五角大廈的武器與監控最後通牒！ (Anthropic AI rejects Pentagon's weapons & surveillance ultimatum)
Uploaded: 2026-02-27T21:43:13.000Z
Duration: 8 min 5 s

Anthropic is rejecting an ultimatum from the Pentagon to lift the company's AI safeguards or risk being blacklisted.

Anthropic 拒絕五角大廈的最後通牒，該通牒要求解除公司的 AI safeguards，否則將面臨被列入黑名單的風險。

Defense Secretary Pete Hegseth wants to use Anthropic's AI model, Claude, for quote, all lawful purposes.

國防部長 Pete Hegseth 希望將 Anthropic 的 AI model，Claude，用於「所有合法目的」。

轉承詞

Anthropic, though, has been very clear it's a hard no when it comes to using its AI technology for autonomous weapons and mass surveillance of Americans.

然而，Anthropic 已經非常明確表示，在將其 AI technology 用於自主武器和對美國人的大規模監控方面，絕對不行。

The deadline's later today, but essentially saying what?

截止日期是今天晚些時候，但基本上是在說什麼？

We are not playing ball with the Pentagon.

Yeah, so they said that the Pentagon came back to them with some updates to this contract they've been negotiating for months.

對，所以他們說五角大廈帶著他們幾個月來一直在談判的這份合約的一些更新又回來了。

And they say that even though they came back with some edits, that the legalese in there still doesn't satisfy their concerns about their two red lines.

而且他們說，儘管五角大廈提出了一些編輯，但其中的法律條款仍然無法滿足他們對兩條紅線的擔憂。

Now, as you know, the Pentagon wants to be able to say to Anthropic, we want to use your Claudia system.

現在，正如你所知，五角大廈希望能夠告訴 Anthropic，我們想使用你們的 Claude system。

That's one of the only systems that's used in the classified military system for any purpose.

這是用於機密軍事系統的少數幾個系統之一，可用於任何目的。

They don't want to be in the middle of a war situation, having to go to Anthropic and saying, hey, we need to do this.

他們不希望在戰爭情況下，還得跑去跟 Anthropic 說，嘿，我們需要做這件事。

But Anthropic says they have two major concerns here, and

但 Anthropic 表示他們在這方面有兩大疑慮，

it's because they just don't think the AI systems are either reliable enough for this.

這是因為他們認為 AI systems 還不夠可靠。

These are things like autonomous weapons.

They say it's just not reliable enough yet to be able to aim and shoot a weapon.

他們說，目前還不夠可靠，無法瞄準和開槍。

They're also concerned about mass surveillance of US citizens, saying just our laws and regulations have not caught up to what AI can do when it comes to mass surveillance.

他們也擔心大規模監控美國公民，認為我們的法律和法規還跟不上 AI 在大規模監控方面的能力。

And they say that the Pentagon's attempted changes did not satisfy that, and

他們說，五角大廈試圖做的修改並未能解決這個問題，

And they say the military is welcome to go elsewhere.

They will help the military off board Anthropic's products.

他們會協助軍方終止使用 Anthropic 的產品。

And Dariel Amadei, the CEO of Anthropic, saying in a statement, regardless, these threats do not change our position.

Anthropic 的 CEO Dariel Amadei 在一份聲明中表示，無論如何，這些威脅都不會改變我們的立場。

We cannot in good conscience accede to their requests.

我們不能在良心上同意他們的請求。

Those threats, of course, being that the Pentagon is going to essentially blacklist Anthropic, calling them a supply chain risk.

當然，這些威脅就是五角大廈將會把 Anthropic 列入黑名單，稱他們是供應鏈風險。

This is something that's normally used for companies that are connected to foreign adversaries.

這通常是用於與外國敵對勢力有關聯的公司。

I don't think it's ever been used on an American AI company, not just any American AI company, but

我認為這從來沒有用在一家美國 AI 公司身上，而且不只是任何一家美國 AI 公司，

one that's considered one of the best in the industry.

而是被認為是業界頂尖的公司之一。

Those Claude systems are loved by programmers.

那些 Claude 系統深受程式設計師喜愛。

There's a reason why the Pentagon is using this because it is well used within the Defense Department.

五角大廈之所以使用，是因為它在國防部內廣泛使用，這是有原因的。

Experts say that it is well within all of the military and the military does enjoy using it.

專家表示，這完全符合軍方使用，而且軍方也很喜歡使用它。

And that's why the Pentagon wants to continue use this and is probably putting up this fight.

這就是為什麼五角大廈想繼續使用它，而且可能正在進行這場鬥爭。

What's really interesting is that the Under Secretary for War on Technology, the one who's been in a lot of these negotiations, Emil Michael, went on a Twitter tear yesterday, posting all of these things about Anthropic, about Dariel Amadei, saying instead of just saying, okay, we're just going to move on to another company, they're accusing Anthropic of trying to rewrite the laws instead of just being a company who wants to uphold its own values in its contract.

比較有趣的是，負責科技戰爭的副國防部長、參與了許多協商的 Emil Michael，昨天在 Twitter 上大發雷霆，發布了許多關於 Anthropic、關於 Dario Amadei 的貼文，說他們沒有隻是說「好，我們要換另一家公司」，而是指控 Anthropic 試圖改寫法律，而不是隻是一家想在合約中堅持自身價值觀的公司。

I'll read you one of the posts that he said.

我來念一段他發的貼文給你們聽。

He said it's a shame that Dario Amadei is a liar and has a God complex.

他說 Dario Amadei 是個騙子，還有個上帝情結。

He wants nothing more than to try to personally control the US military military and is okay putting our nation's safety at risk.

他只想個人控制美國軍方，而且不惜拿國家安全冒險。

The Department of War will always adhere to the law, but

not bend to the whims of any one for profit tech company.

但不會屈服於任何一家圖利的科技公司的意願。

And again, you're thinking, well, if you just don't like this AI company, just go use something else.

再說一次，你可能會想，嗯，如果你不喜歡這家 AI 公司，那就換別家就好。

And part of the reason is Anthropic is one of the only AI systems that's approved to work on the classified military systems.

部分原因在於，Anthropic 是少數獲準在機密軍事系統上運作的 AI 系統之一。

The other AI companies are in talks to do so as well.

for some people, they say none of them are as good as Claude and that's why you have this fight playing out.

有些人說，沒有一家 AI 公司比 Claude 還好，這就是你們看到的爭論。

So we got the statement from Anthropic last night.

所以我們昨晚收到了 Anthropic 的聲明。

The deadline is Technically Today, 5:01pm I believe.

我認為截止日期是今天下午 5:01。

What do you make of the response from Anthropic?

你覺得 Anthropic 的回應怎麼樣？

I think the response from Anthropic demonstrates something that's been clear from the start, which is they really want to work with it, with the Pentagon.

我認為 Anthropic 的回應展現了從一開始就很清楚的一點，那就是他們真的很想和五角大廈合作。

Anthropic was the first company in the door of the frontier AI labs to do classified work.

Anthropic 是第一家進入前沿 AI 實驗室並從事機密工作的公司。

And one of the really interesting things about this, this whole dispute is that every single thing that the Pentagon has wanted Anthropic to work on, Anthropic has been happy to work on.

而這場爭議中一個非常有趣的地方是，五角大廈希望 Anthropic 做的每一件事，Anthropic 都樂意去做。

And even the statement last night from, from Anthropic leaders suggested that they're happy to work on something like autonomous weapons.

而且昨晚 Anthropic 領導人的聲明甚至暗示，他們樂意從事像是自主武器這類的工作。

They just don't think their technology is ready, is ready for it, ready for it right now.

他們只是覺得他們的技術還沒準備好，還沒準備好，還沒準備好。

They're saying, to your point, we will work on this.

他們說，就像你說的，我們會處理這個。

And as my colleague had asked Gold Our correspondent here was reporting, look, folks in the military, this is seen as the best AI option for them right now.

我的同事，在這裡的記者 Gold，他有報導說，軍方認為這是他們目前最好的 AI 選擇。

They would have to then basically off board everything that they already have, bring in something else.

他們就得把現有的東西全部換掉，換成別的東西。

It's interesting that if there is a solution that seemed to somewhat be working and the company is saying, hey, our technology is not even there yet to do this, it would seem that from a national security standpoint, you wouldn't want to use something that wasn't ready for primetime.

有趣的是，如果有一個解決方案似乎有點奏效，而公司卻說，嘿，我們的技術還沒到能做這個的程度，從國家安全的角度來看，你應該不會想用還沒準備好的東西吧。

One of the interesting things here is this feels to me like a personality dispute dressed up as a policy disagreement in that there's no current thing that the Pentagon wants Anthropic to work on that Anthropic is not willing to work on.

有趣的一點是，這在我看來，像是把個人恩怨包裝成政策分歧，因為五角大廈現在沒有任何想讓 Anthropic 處理的事情是 Anthropic 不願意做的。

And that suggests to me that this in some ways is a theoretical dispute about future potential use cases.

這也暗示著，這在某種程度上是關於未來潛在用途的理論性爭論。

And really what this reflects is a breakdown in trust in that the Pentagon doesn't trust that Anthropic will be there for future potential uses where it might want to contract with a Frontier AI lab.

這其實反映了信任的破裂，因為五角大廈不相信 Anthropic 在未來可能需要與前沿 AI 實驗室簽約時會出現。

And Anthropic, for its part, doesn't seem to trust that the Pentagon will necessarily use its technology responsibly.

而 Anthropic 方面，似乎也不相信五角大廈會負責任地使用他們的技術。

This is something that they should be able to work out, given that both sides seem to be benefiting from the arrangement at present.

考量到雙方目前似乎都從這次合作中獲益，這應該是他們可以解決的問題。

Anthropic, correct me from our here, Anthropic has also been pushing for some guardrails when it comes to AI technology, which has not necessarily been where the administration stands.

Anthropic，如果我說錯了請糾正我，Anthropic 一直在推動 AI safeguards，這和政府的立場不一定一致。

Do you think that could be playing a role here?

你覺得這可能在其中扮演了什麼角色嗎？

I think Anthropic publicly has been very.

我認為 Anthropic 在公開場合一直非常。

They're the frontier lab most known for their stance on on air safety and have been very vocal about these guardrails against mass surveillance and autonomous weapon systems.

他們是最知名的前沿實驗室，以其在 AI safety 方面的立場聞名，並一直大力倡導 AI safeguards 來防止大規模監控和自主武器系統。

But one of the really interesting things is, you know, from the statement by Anthropic last night, it appears as though their objections are less philosophical and more practical.

但其中一個非常有趣的地方是，你知道，從 Anthropic 昨晚的聲明來看，他們的異議似乎不是哲學上的，而是實際上的。

Technology isn't ready for this kind of task yet.

技術還沒準備好應付這種任務。

And frankly, for example, when it comes to autonomous weapons, they're not wrong.

說實話，舉例來說，就自主武器而言，他們說的沒錯。

If you were going to train an autonomous weapons system today, you'd probably be better off with a bespoke algorithm trained on a data set of whatever the specific kinds of targets you were trying to hit, rather than a large language model trained on the Internet or something like that.

如果你今天要訓練一個自主武器系統，你可能會比較適合使用一個專門演算法，該演算法是在針對你想要攻擊的特定目標的資料集上訓練出來的，而不是使用一個在網路上訓練出來的大型語言模型之類的。

What Anthropic is saying is that our technology isn't ready yet to do certain things and we need guardrails against using it for those things until it's ready and the Pentagon is saying, we want those guardrails down now.

Anthropic 的說法是，我們的技術還沒準備好做某些事情，我們需要有防護措施來防止在技術準備好之前被用於這些用途，而五角大廈則說，我們現在就要移除這些防護措施。

And we find the idea that you might have guardrails to be the problem in the first place because you're doing business with the Pentagon.

我們認為，因為你們和五角大廈有生意往來，所以有防護措施本身就是個問題。

And to be fair to the Pentagon here, Anthropic is in business with the Pentagon and the Pentagon is in the business of war.

說句公道話，Anthropic 和五角大廈有生意往來，而五角大廈的業務就是戰爭。

And so we can't be surprised in some ways that the Pentagon wants access to technology from one of its contractors to do the things that it wishes to do to try to improve the American military.

所以，從某些方面來看，我們不能驚訝於五角大廈希望從其承包商之一那裡獲得技術，以便做他們想做的事情來改善美國軍隊。

Real quickly, before I let you go, for people who are not following every twist and turn here, when you hear Pentagon and mass surveillance, I think it does raise some issues for average Americans.

在我讓您離開之前，快速說一下，對於那些沒有緊追時事發展的人來說，當聽到五角大廈和大規模監控時，我認為這確實會讓一般美國人產生一些疑慮。

How concerned should Americans be about mass surveillance from the Department of Defense on them, the average citizen?

美國人應該對國防部對他們，也就是一般民眾進行大規模監控有多擔心？

I think people have reasons to be concerned about mass surveillance, though it's not clear to me that that concern comes from the Pentagon per se.

我認為人們有理由擔心大規模監控，不過我並不確定這種擔憂是否來自五角大廈本身。

The public statements from the Pentagon have suggested that they're not doing mass surveillance, which is one reason they think that anthropics, that anthropics, guardrails here are not necessary, reflecting that lack of trust between Anthropic and the Pentagon I mentioned before.

五角大廈的公開聲明表示他們並未進行大規模監控，這也是他們認為這裡的 anthropics、anthropics 的 guardrails 並非必要的原因，這反映了我之前提到的 Anthropic 和五角大廈之間缺乏信任。

But yeah, I would be worried about mass surveillance in an age of this technology.

但沒錯，在這個 AI 科技的時代，我會擔心大規模監控。

The Pentagon is not necessarily the locust point I'd be most worried about.

五角大廈不一定是讓我覺得最擔憂的焦點。