Name: Anthropic 與五角大廈的爭議背後有什麼？ | DW News (What's behind the Anthropic-Pentagon dispute? | DW News)
Uploaded: 2026-02-25T23:08:03.000Z
Duration: 5 min 46 s

AI Co. Anthropic is getting under pressure from the Pentagon.

AI 公司 Anthropic 正受到五角大廈的壓力。

Defense Secretary Pete Hegseth has given the AI company an ultimatum until Friday to agree to give the military full access to the company's AI models or face cuts of up to $200 million in government contracts.

國防部長 Pete Hegseth 已給予這家 AI 公司最後通牒，要求在週五前同意讓軍方全面存取該公司的 AI 模型，否則將面臨高達 2 億美元的政府合約被削減。

情態副詞

Hegseth reportedly demanded the AI company drop its safeguards.

據報導，Hegseth 要求這家 AI 公司放棄其安全措施。

Anthropic, a breakaway from OpenAI, has restricted military use of its models.

Anthropic，一家從 OpenAI 分拆出來的公司，已限制軍方使用其模型。

That includes autonomous lethal targeting and surveillance.

這包括自主致命目標鎖定和監控。

The Pentagon is considering invoking the Defense Production act, which could force Anthropic to allow unrestricted military use on national security grounds.

五角大廈正考慮引用國防生產法案，該法案可能迫使 Anthropic 允許在國家安全理由下無限制的軍方使用。

Let's bring in Vanessa Voels, who's a researcher at the Bundeswehr University in Munich.

我們來連線慕尼黑聯邦國防軍大學的研究員 Vanessa Voels。

The Pentagon wants to use AI for what it's calling, quote, unquote, all lawful purposes.

五角大廈希望將 AI 用於他們所謂的「所有合法用途」。

From a defense perspective, what does that, or could that actually mean in practice?

從國防角度來看，這實際上意味著什麼，或者說可能意味著什麼？

So it seems like the Pentagon wants to go beyond the actually agreed terms from last year, which was set not to use autonomous weapons systems without human control.

所以看起來五角大廈想超越去年達成的協議條款，也就是不使用沒有人類控制的自主武器系統。

So basically it seems that now it wants to use those technologies that go beyond ISR purposes, so not only for intelligence, surveillance and reconnaissance purposes, but

所以基本上現在似乎想使用那些超越 ISR 目的的技術，不只是用於情報、監控和偵察，而是

So maybe this could lead to selecting and engaging targets with drones across all military domains and eventually having no human operator to control that.

所以這可能會導致在所有軍事領域，利用無人機選擇和攻擊目標，最終沒有人類操作員來控制。

However, it is long standing US Policy to only use autonomy weapon systems with quote, appropriate levels of human judgment.

然而，美國長期以來的政策是，只能在「適當程度的人類判斷」下使用自主武器系統。

Therefore, we should not expect like Terminator scenarios where machines are fighting machines, but

是以，我們不應該期待像魔鬼終結者那樣機器與機器互相戰鬥的場景，但

certainly to enhance and to enhance those technologies going more towards the kinetic end.

肯定會增強這些技術，使其朝向實際軍事應用發展。

Can you tell us a little bit more about the military use of artificial intelligence for battlefield applications?

你能多告訴我們一些關於人工智慧在戰場應用上的軍事用途嗎？

You know, you've mentioned there we're talking about spanning autonomous drone swarms, robots, cyber attacks, for example.

你知道，你提到我們正在談論涵蓋自主無人機群、機器人、網路攻擊等等。

You know, what are, what are the big concerns here and why does Anthropic have these safeguards in particular?

說到這個，大家最關心的問題是什麼？為什麼 Anthropic 特別會有這些安全措施呢？

So basically when we talk about military purposes, we obviously want the armed forces to control their own technologies and their own systems.

所以基本上，當我們談論軍事用途時，我們當然希望軍方能掌控自己的技術和系統。

When it comes to AI, they might sort of develop a life that goes beyond what was programmed and what was intended.

說到 AI，它們可能會發展出不受控制的生命，超越了原本的程式設定和預期。

Therefore, we have long standing international discussions on regulations.

是以，我們長久以來一直在進行關於監管的國際討論。

There's no international treaty yet, but we certainly want to make sure that we have appropriate human control.

雖然還沒有國際條約，但我們當然希望能確保有適當的人類控制。

Therefore, we don't want the machine to decide and select on targets, but

所以，我們不希望機器來決定和選擇目標，但是

我們還是希望軍方來做這件事。

And that's what is being discussed here and which is developing not only in democratic states, but

這就是這裡正在討論的，而且不僅在民主國家，

Of course, we are hearing that the Pentagon is negotiating with other large language models.

當然，我們聽說五角大廈正在和其他大型語言模型洽談。

And through that lens I'd like to ask you, because I mean Anthropic, it is holding firm right now to its red lines around autonomous weapons and mass surveillance.

從這個角度來看，我想問你，因為我的意思是 Anthropic，它現在堅守在自主武器和大規模監控的紅線。

I'm thinking about Google, OpenAI XAI, what is their position on military use cases in comparison?

Google、OpenAI、XAI，他們對於軍事用途的立場又是如何？

So the DoD signed contracts with all these companies last year and summer where they also agreed on certain red lines, as we mentioned before.

所以國防部去年夏天和所有這些公司簽訂了合約，正如我們之前提到的，他們也同意了一些紅線。

However, now there seems to be a new development, but

然而，現在似乎有了新的發展，但是

so far only Anthropic is being threatened with being listed as a supply chain risk and the others are not.

到目前為止，只有 Anthropic 被威脅要列為供應鏈風險，其他公司則沒有。

So this seems to lead to the conclusion that only Anthropic is really focusing on these safeguards.

所以這似乎得出一個結論，只有 Anthropic 真的在關注這些安全措施。

And that makes sense because in the early days of Anthropic, they were actually breaking away from OpenAI in order to put a bigger focus on safeguarding safety, reliability.

這也說得通，因為在 Anthropic 的早期，他們實際上是從 OpenAI 分拆出來的，以便更專注於保障安全、可靠性。

Therefore, it seems that Anthropic is leading in this regard and this is important for the DoD because so far this is actually the only company that has been included in classified U.S.

是以，看來 Anthropic 在這方面處於領先地位，這對國防部來說很重要，因為到目前為止，這實際上是唯一一家被納入美國機密的

So there is a big competition between those companies going on.

所以這些公司之間正在展開一場激烈的競爭。

You know, the Pentagon is pushing this issue very aggressively.

你知道，五角大廈正非常積極地推動這個議題。

They've set this deadline now on Friday and I'd like to ask you, what do you think happens after that deadline expires and what are you thinking, you know, as an expert in this space, looking at what is, what is unfolding right here.

他們現在設定了週五的最後期限，我想問你，你認為過了這個期限之後會發生什麼事？而且，作為這個領域的專家，你看到現在正在發生的事情，你又是怎麼想的呢？

So in order to understand Anthropic, we need to understand that this idea of safety and reliability was really the founding principle of the company.

所以，為了了解 Anthropic，我們需要明白，安全和可靠性的理念確實是這家公司的創立宗旨。

So personally I don't think that they can actually go beyond those self set red lines.

所以我個人認為，他們實際上無法超越自己設定的紅線。

Therefore I think think that they need to have certainly more discussions and will not directly follow this pressure by the dod.

是以，我認為他們肯定需要有更多的討論，而且不會直接屈服於國防部的壓力。

Vanessa Vaux with the Bundeswehr University in Munich.

來自慕尼黑聯邦國防軍大學的 Vanessa Vaux。

Unfortunately, we have to leave it there.

I could ask you many more questions about this topic, of course, but

當然，關於這個話題我還有很多問題想問你，但是

字幕列表影片播放

Anthropic 與五角大廈的爭議背後有什麼？ | DW News (What's behind the Anthropic-Pentagon dispute? | DW News)

perspective

engage

basically

force