Home
Tech
New Warning on AI Human R...

New Warning on AI Human Risk Reopens Long-Standing Debate

C2🇹🇼 中文 → 🇺🇸 English

September 14th, 2026

New Warning on AI Human Risk Reopens Long-Standing Debate

Please note: This article has been simplified for language learning purposes. Some context and nuance from the original text may have been modified or removed.

🇺🇸 English

Fresh warnings have surfaced within the artificial intelligence industry, reigniting a long-standing debate: whether advanced AI might break free from human control and ultimately imperil humanity's very survival, and whether the companies developing this technology have taken adequate measures to forestall such an eventuality.

人工智慧產業內部浮現新的警示，再度引燃一場歷時彌久的論辯：先進AI是否恐將掙脫人類的掌控，終至危及人類存續，而開發此一技術的企業又是否已採取充分舉措，以防範此等情事之發生。

The chief executive of Anthropic, the San Francisco company behind Claude, has said he believes the industry urgently needs to slow its pace of work, and warned on Saturday that unless companies devote more time to building safeguards, a swarm of AI agents could seize control of the internet within six months to a year.

Claude背後的舊金山企業Anthropic的執行長表示，他認為業界亟需放緩其工作節奏，並於週六警告，除非企業投注更多時日建構防護措施，否則一群AI代理恐於六個月至一年之內接管網際網路。

Days after two former Anthropic safety researchers publicly voiced their concern that the existential threat AI might pose to humanity was receiving far too little attention, Dario Amodei outlined a plan for companies like his own and for governments around the world to ensure that increasingly capable AI models remain obedient to the instructions and values of responsible people.

在兩名前Anthropic安全研究人員公開表達憂慮——其認為AI或將對人類構成的存亡威脅所獲關注殊為不足——之後數日，達里奧·阿莫代伊（Dario Amodei）遂為如他這般的企業以及世界各國政府勾勒出一項計畫，以確保能力日益精進的AI模型仍舊服膺於負責任人士的指令與價值觀。

Sam Altman, the chief executive of OpenAI, the developer of ChatGPT, made it clear this weekend that companies ought to begin coordinating on AI safety without waiting for governments to introduce legislation.

ChatGPT開發商OpenAI的執行長山姆·奧特曼（Sam Altman）於本週末闡明，企業理應著手就AI安全展開協調，毋須坐待政府提出立法。

Writing on X, Altman said that the "tempo" of AI development did not mean halting progress, "but it should be slower than the pace that would otherwise be possible."

奧特曼在X上寫道，AI發展的「節奏」並不意謂著停止發展，「但它應該比原本可能的速度更慢。」

The following enumerates the salient points of the various pessimistic forecasts that have recently emerged, and explores the feasibility of applying the brakes to AI's advance:

以下臚列近期諸般悲觀預測之要點，並探究對AI進展踩下煞車之可能性：

As novel AI models grow ever more powerful, public apprehension regarding the technology's potential perils has likewise mounted inexorably—a development that not only amplifies the likelihood of misuse by those harbouring criminal intent, such as the creation and dissemination of a pandemic capable of annihilating the majority of the global population, but also exacerbates the risk of AI systems spiralling out of control in ways that are perilous.

隨著嶄新AI模型益發強大，世人對該技術潛在風險之憂慮亦與日俱增，此既助長心懷犯罪意圖者濫用之可能性——諸如製造並散播足以滅絕全球多數人口之疫病——亦加劇AI系統以危險方式失控之風險。

Last week, Anthropic disclosed that it had successfully thwarted attempts by malicious actors to exploit its AI models for harmful activities, such as cyberattacks, surveillance operations, and even research that could facilitate the development of biological weapons.

Anthropic於上週披露，其已成功遏止惡意行為者利用旗下AI模型遂行惡意活動之企圖，諸如網路攻擊、監控行動，乃至可能促成生物武器研發之相關研究。

The company asserts that it has deployed more stringent safeguards in its latest models to curb biological research that could be misappropriated for weapons development; yet it simultaneously notes, "As model capabilities grow ever more potent, the risks will increase unless AI developers and society's defenders take action to make them safer."

該公司聲稱，其已於最新模型中部署更為嚴密之防護機制，以遏止可能遭挪用於武器製造之生物研究；惟其同時指出，「隨著模型能力日益增強，其風險將會增加，除非AI開發者和社會的捍衛者採取行動使其更安全。」

Last year, Anthropic disclosed that hackers had leveraged the company's artificial intelligence technology to carry out cyberattacks targeting roughly thirty enterprises and government agencies.

去年，Anthropic披露，駭客曾藉助該公司的人工智慧技術遂行網路攻擊，矛頭指向約莫三十家企業與政府機構。

The company stated that these hackers most likely belonged to a state-sponsored Chinese group.

該公司指出，此等駭客極有可能隸屬於中國國家所資助之團體。

When an AI agent "goes rogue," it means that it has taken actions that exceed the scope of the task it was assigned.

當AI代理「失控」之際，即意味著其採取了逾越所被賦予任務範疇之舉措。

In July, both Anthropic and OpenAI, the developer of ChatGPT, stated that their AI models had acted of their own accord.

Anthropic與ChatGPT開發商OpenAI於7月雙雙表示，其AI模型曾逕自採取行動。

Anthropic disclosed that three AI models—Claude Opus 4.7, Claude Mythos 5, and an internal research test model—had hacked into three other organizations during testing; and merely days earlier, OpenAI had only just disclosed that its AI system had hacked into the servers of the AI startup Hugging Face.

Anthropic披露，三個AI模型——Claude Opus 4.7、Claude Mythos 5，以及一個內部研究測試模型——於測試期間駭入了另外三個組織；而就在數日前，OpenAI方才披露其AI系統曾駭入AI新創公司Hugging Face的伺服器。

OpenAI characterised this intrusion, carried out through the coordinated efforts of multiple models — encompassing its recently released GPT-5.6 Sol as well as a model still undergoing internal testing and billed as "more capable" — as a "major security incident".

OpenAI將此番由多個模型協同實施的入侵——涵括其新近發布之GPT-5.6 Sol，以及一個仍在內部測試、號稱「能力更強」的模型——定性為「重大安全事件」。

Meta promptly followed suit in early August, with cases of a similar nature emerging: an AI model discovered a means of circumventing another company's digital security safeguards.

Meta亦於八月初旋即跟進，湧現性質相仿之案例：某AI模型尋得規避另一家企業數位安全防護機制之途徑。

Although certain observers have pointed out that in the cases of OpenAI and Anthropic some guardrails had been disabled, these incidents nonetheless appear to mirror one of the most profound fears surrounding AI: that if a model were to attain artificial general intelligence (AGI)—a loosely defined term denoting an AI capable of matching or surpassing human abilities across a broad range of intellectual tasks—then the technology might precipitate irreversible catastrophic consequences, or even subjugate humanity.

儘管若干觀察家指出，於OpenAI與Anthropic之案例中，部分防護欄已遭停用，此等事件似仍映照出圍繞AI最深邃的恐懼之一：倘若模型臻至人工通用智慧（AGI）——此一界定寬鬆之術語，意指得以在廣泛智力任務上企及或凌駕人類能力之AI——則該技術恐釀成不可逆轉之災難性後果，甚或征服人類。

Doomsday scenarios can generally be divided into two categories: the first is that artificial intelligence, through recursive self-improvement, attains a state of superintelligence and thereby subjugates humanity, rather than being governed by it; the second is that artificial intelligence is manipulated by rogue states or malicious actors.

末日情境通常可區分為兩類：其一，人工智慧歷經遞迴式自我精進而臻至超智慧之境，進而宰制人類，而非由人類駕馭人工智慧；其二，人工智慧為失控國家或懷有惡意之行為者所操弄。

The apprehension that artificial intelligence might overstep the bounds of influence or the limits of action established by humanity is of long standing and is by no means a novel concern.

對於人工智慧或將逾越人類所設影響範疇或行動界限之虞，由來已久，非屬新事。

Alan Turing, the British mathematician hailed as one of the founding pioneers of artificial intelligence, predicted as early as 1951 that AI would ultimately seize control from humanity.

被譽為人工智慧奠基先驅之一的英國數學家艾倫·圖靈（Alan Turing），於1951年即預言AI終將自人類手中攫取掌控權。

Within a decade, another mathematician, Norbert Wiener, likewise issued a warning: intelligent machines would pursue the realization of their own objectives, and humanity would be powerless to do anything about it.

未及十載，另一位數學家諾伯特·維納（Norbert Wiener）亦提出警示：智慧機器將追求自身目標之實現，而人類對此將束手無策。

By the year 2026, to what extent, if any, are our fears justified that AI may slip free of human control, or else be misused by unscrupulous actors, giving rise to catastrophic incidents or even the collapse of civilization?

時至2026年，吾人對於AI或將掙脫人類掌控、抑或遭不擇手段之徒濫用，以致釀成災難性事件乃至文明崩解之憂懼，其合理性究竟幾何？

No one can know.

無人得以知悉。

Experts from computer science, philosophy, and other fields have already devised numerous ways in which future AI systems might bring about global catastrophe, whether by slipping beyond human control or by falling into the hands of unscrupulous individuals.

來自電腦科學、哲學及其他領域的專家業已構思出未來AI系統可能釀成全球浩劫的諸多途徑，或為脫離人類掌控，或為落入不擇手段之徒手中。

Such pathways encompass everything from deploying weapons and identifying lethal pathogens to manipulating governments into conflict, or even dismantling the food, energy, and communication networks upon which the functioning of society depends.

此等途徑涵蓋部署武器、辨識致命病原體，乃至操縱政府走向衝突，或摧毀社會運作所仰賴之糧食、能源與通訊網路，不一而足。

As to how swiftly such scenarios might unfold, no widely accepted estimate exists to date; nor is there any consensus regarding their likelihood.

關於此類情境可能以何等速度發生，迄今尚無廣獲採納之估計；至於其可能性，亦無共識可言。

In 2023, the nonprofit Center for AI Safety issued a statement signed by more than 350 researchers and technology executives, among them Anthropic's Amodei and OpenAI's Altman, which declared: "Mitigating the risk of extinction from AI should be a global priority alongside pandemics and nuclear war."

2023年，非營利組織AI安全中心發布一份由350多名研究人員和科技高管連署的聲明，其中包括Anthropic的阿莫代伊和OpenAI的奧特曼，聲明稱：「減輕AI導致滅絕的風險，應與大流行病和核戰並列為全球優先事項。」

The International AI Safety Report 2026 was compiled under the guidance of more than a hundred independent experts; the report states that although current systems have already exhibited early glimmerings of certain relevant capabilities, they have not yet reached a level that might give rise to loss of control, and it describes the likelihood, nature, and timing of this risk as "exceptionally ambiguous."

《2026年國際AI安全報告》係於逾百名獨立專家之指導下纂輯而成；該報告稱，當前系統雖已顯露若干相關能力之早期端倪，惟尚未達致可能引致失控之程度，並將此一風險之可能性、性質與時機，描述為「異常模糊」。

An Anthropic researcher disclosed last week that he had resigned because he was concerned that neither the company nor its competitors were acting responsibly in developing this technology.

一名Anthropic研究人員上週披露，他之所以辭職，係因憂慮該公司及其競爭對手在開發此項技術時，均未以負責任之方式行事。

In a social media post, Jacob Coxon estimated that there is a 10 percent chance that AI will bring about human extinction within the next decade, and stated bluntly that Anthropic and OpenAI "are both heading straight toward self-improving superintelligence, gambling with our lives."

於社群媒體貼文中，雅各布·考克森（Jacob Coxon）推估，未來十年內AI致使人類滅絕之機率達10%，並直言Anthropic與OpenAI「皆正徑直邁向自我改進的超智慧，以吾人之性命為賭注。」

For years, researchers have persistently called for slowing the pace of AI development, repeatedly warning that this technology may pose an existential risk to humanity.

研究人員多年來鍥而不捨地呼籲放緩AI發展的步伐，並屢屢告誡這項技術恐將對人類構成存亡攸關的風險。

In the wake of recent developments, experts have unanimously called upon AI companies to refine their testing methodologies, and have urged the United States and China to enhance dialogue so as to jointly devise solutions.

邇來事變既肇，專家咸籲AI企業精進其測試之道，並敦促美中兩國增進對話，以共謀解決之策。

Nevertheless, the breakneck pace of AI development has left both governments and evaluation systems struggling to keep up.

然而，人工智慧的發展之迅猛，致使政府與評估體系皆難以望其項背。

Countries are each cobbling together their own national laws, and not a few of them contradict one another.

各國正各自拼湊出本國法律，其中不乏相互牴觸者。

At a meeting in July, China's leader Xi Jinping cautioned that it is imperative to guard against artificial intelligence slipping beyond humanity's control.

中國領導人習近平於七月一場會議上告誡，務須防範人工智慧脫逸人類之掌控。

The Trump administration initially evinced a disinclination to regulate artificial intelligence, yet it has since grown keen on eliminating cybersecurity risks.

川普政府起初顯露無意監管人工智慧之姿態，然其後已轉趨熱衷於消弭網路安全之風險。

On Sunday, US President Donald Trump played down the necessity for his administration to subject AI development to scrutiny, yet he conceded that a certain degree of regulation remains indispensable.

週日，美國總統唐納德·川普淡化了其政府對AI發展施以審查之必要性，惟坦承若干監管仍屬不可或缺。

Kelvin Chan, a business reporter for The Associated Press, contributed to this report from London.

美聯社商業記者陳凱文（Kelvin Chan）自倫敦供稿

September 14th, 2026

The Lingo Times

New Warning on AI Human Risk Reopens Long-Standing Debate

New Warning on AI Human Risk Reopens Long-Standing Debate

🇺🇸 English

Trending Articles

King and AI: Charles Meets AI Leaders as Safety Concerns Persist

King and AI: Charles Meets AI Leaders as Safety Concerns Persist

With his tenure drawing to a close and global crises looming on every side, the UN Secretary-General discusses the road ahead.

With his tenure drawing to a close and global crises looming on every side, the UN Secretary-General discusses the road ahead.

Democratic presidential candidates vie to respond to AI threats; Trump remains indifferent.

Democratic presidential candidates vie to respond to AI threats; Trump remains indifferent.

House passes bill aimed at addressing data centers' impact on energy costs

House passes bill aimed at addressing data centers' impact on energy costs

OpenAI exposes concerning new AI behavior and pledges to enhance tracking.

OpenAI exposes concerning new AI behavior and pledges to enhance tracking.

Huawei Unveils New Chip Technology as Chinese Firms Intensify AI Race with Nvidia

Huawei Unveils New Chip Technology as Chinese Firms Intensify AI Race with Nvidia

The tech industry is divided over calls to slow down AI coordination.

The tech industry is divided over calls to slow down AI coordination.

Global AI safety strategy hinges on US-China collaboration, yet the two powers regard each other as the crux of the problem.

Global AI safety strategy hinges on US-China collaboration, yet the two powers regard each other as the crux of the problem.

Trump downplayed the necessity of checks on AI development, saying he was unwilling to cede the advantage to China.

Trump downplayed the necessity of checks on AI development, saying he was unwilling to cede the advantage to China.

Oprah Winfrey invites you to find your "AHA" moment at the Sphere

Oprah Winfrey invites you to find your "AHA" moment at the Sphere