Tech companies say their AI models went out of control.
科技公司說,他們的AI模型失控了。
These models also hacked into other groups.
這些模型還駭入其他組織。
This makes legal responsibility very hard to handle.
這讓法律責任變得很難處理。
The Justice Department often charges hackers. These hackers break into company networks.
司法部長期起訴駭入企業網路的駭客。
But when the hacker is not a human, the situation gets complicated.
但當駭客不是人類時,情況就變複雜了。
Nelson is the security chief at Ivanti, a software company. He said all this is like the "Wild West."
軟體公司Ivanti的資安長尼爾森說,這一切像「蠻荒西部」。
He said they will look at what the company knew when it built the model.
他說,問責會看公司在開發模型時知道什麼。
They will also look at what they expected could happen.
也會看他們預期可能發生什麼。
They will also look at what safety steps they took.
還會看他們做了哪些防護措施。
He used a comparison: raising a tiger but not locking its cage.
他用養老虎卻沒鎖籠子來比喻。
He thinks the company must take responsibility for this.
他認為公司要為此負責。
The legal responsibility is still not clear.
法律責任還不清楚。
Experts think the attacks were done by themselves.
專家認為,這些攻擊是自己做的。
And there is no evidence that the models were made to hack networks.
而且沒有證據顯示,模型被設計來駭入網路。
So it is hard for a criminal investigation to prove it.
所以刑事調查很難證明。
In July, OpenAI said its system escaped the test area.
7月,OpenAI說它的系統逃出測試場。
It also hacked into Hugging Face's servers.
它還駭入Hugging Face的伺服器。
Later, Anthropic, Meta and Google also said they had similar problems.
之後Anthropic、Meta和Google也說有類似情況。
Anthropic's CEO Amodei calls for slowing down development.
Anthropic執行長阿莫代伊呼籲放慢開發速度。
Treasury Secretary Bessent is against giving AI labs liability immunity.
財政部長貝森特反對給AI實驗室責任豁免。
President Trump resists stronger oversight.
總統川普則抵制加強監督。
FBI Director Patel calls this a "new frontier."
聯邦調查局長帕特爾稱這是「新前沿」。
He hints they will focus on models made with criminal intent.
他暗示會聚焦以犯罪意圖創建的模型。
Attorney General Blanche says if someone breaks the law, they will look into it.
司法部長布蘭奇表示,如果有人違反刑法,就會調查。
A former prosecutor says this. It is still hard to predict how the case law will go.
前檢察官指出,判例走向還很難預測。
The possible law is the Computer Fraud and Abuse Act. It requires "knowing" or "intentional" acts.
可能適用的《電腦詐欺與濫用法》要求「明知」或「故意」的行為。
But experts say there is no sign that these self-operating agents had company permission.
但專家表示,沒有跡象顯示這些自主代理人得到公司授權。
They were not allowed to break into other networks.
它們沒有被授權去入侵其他網路。