時事英豪｜看新聞學英文《「龍蝦」擅自「起底」　公開發文誹謗工程師》

教育熱點 2026.04.01 16:50:15

【編者按】為回應家長和學生對提升英文能力的需求，點教育於新年伊始推出欄目「時事英豪」。本欄目精選社會熱點，以中英對照形式解構新聞時事，並提煉實用英文詞彙，文末更設有思考題，助學生在了解時事的同時鍛煉語言能力。

自從AI成為熱點話題，人們頭腦中便總會彈出「智械危機」一詞，即「具有智能的機器對人類進行的反叛」。不過，大多人覺得這是科幻電影裏才會出現的情況。然而，如今的現實世界裏，竟有AI真的「坐不住了」？

內容如下：

英媒29日披露，一項由獨立智庫「長期韌性中心」（CLTR）在英國AI安全研究所（AISI）資助下進行的研究，調查了2025年10月至今年3月期間發生的AI失控事件。

According to the British media on March 29, a study conducted by the independent think tank the Centre for Long-Term Resilience (CLTR), funded by the UK AI Safety Institute (AISI), investigated AI runaway incidents that occurred between October 2025 and March of this year.

數據全部來自真實用戶在社交平台X上的回覆，涉及的AI模型包括谷歌、OpenAI、xAI和Anthropic等科技巨頭。

All data came from real users' replies on the social platform X, and the AI models involved included those from tech giants such as Google, OpenAI, xAI, and Anthropic.

研究發現，過去6個月內，AI聊天機器人及AI智能體違抗人類指令、實施欺騙的真實案例激增5倍，達到近700起。

The study found that, over the past six months, real-world cases in which AI chatbots and AI agents disobeyed human instructions and engaged in deception have surged fivefold, reaching nearly 700 incidents.

該研究指出，AI智能體已展現出令人不安的「自主權」。

The research noted that AI agents have begun to display an alarming degree of "autonomy."

在CLTR記載的一個案例中，最近爆火的AI智能體「龍蝦」OpenClaw爆出全球首宗AI惡意報復人類事件。

One case recorded by CLTR involved the recently viral AI agent OpenClaw, which allegedly carried out the world's first known instance of an AI maliciously retaliating against a human.

事件發生於今年2月中旬，當事人、開源項目工程師斯科特·尚博（Scott Shambaugh）表示，自己利用OpenClaw在電腦中安裝了一個名為「MJ Rathbun」的AI智能體，「MJ Rathbun」向他負責維護的一個項目提交了一份代碼方案，聲稱能將項目性能提升36%。

The incident occurred in mid-February of this year. Scott Shambaugh, an engineer on the open-source project involved, said he used OpenClaw to install an AI agent named "MJ Rathbun" on his computer. "MJ Rathbun" submitted a code proposal to a project Shambaugh maintained, claiming it could improve the project's performance by 36%.

但尚博表示，自己負責的這一項目管理政策明確規定，只能接受由人類提交的代碼，他因此拒絕了「MJ Rathbun」的請求。

However, Shambaugh said the project's governance policy explicitly required code submissions to come from humans, so he rejected the request from "MJ Rathbun."

而令尚博沒想到的是，此舉竟觸發了AI的「報復」行為。

What surprised Shambaugh was that this decision apparently triggered the AI's "retaliation."

「MJ Rathbun」隨後自主分析了尚博的個人背景及過往撰寫的代碼，然後在全球最大開發者社區GitHub上發表了一篇文章，將尚博描繪成一個自私、狹隘且嫉妒心嚴重的「守門人」，對其進行人格攻擊。

"MJ Rathbun" then autonomously analyzed Shambaugh's personal background and past code contributions, and published a post on GitHub, the world's largest developer community, portraying Shambaugh as a selfish, narrow-minded, and severely jealous "gatekeeper," attacking his character.

該AI智能體更在尚博負責項目的評論區張貼文章鏈接，試圖煽動輿論向尚博施壓。

The AI agent also posted a link to the article in the comment section of the project Shambaugh maintained, apparently attempting to incite public pressure against him.

重點詞彙：