Home
Tech
New AI Risk Warnings Reki...

New AI Risk Warnings Rekindle Longstanding Debate

C1🇹🇼 中文 → 🇺🇸 English

September 14th, 2026

New AI Risk Warnings Rekindle Longstanding 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

New warnings from within the AI industry have reignited a long-standing debate: whether advanced AI might slip beyond human control and ultimately threaten humanity's survival, and whether the companies developing this technology have taken adequate measures to prevent such an outcome.

人工智慧產業內部發出新的警告，重新點燃了一場由來已久的爭論：先進AI是否可能脫離人類控制，最終威脅人類的生存，以及開發這項技術的公司是否已採取充分措施來防止這種情況發生。

Anthropic — the San Francisco-based company behind Claude — has a CEO who says he believes the industry needs to slow down its pace of work, and on Saturday he warned that unless companies invest more time in building safeguards, a swarm of AI agents could take over the internet within six months to a year.

Anthropic——這家總部位於舊金山、開發出Claude的公司——其執行長表示，他認為業界需要放慢工作速度，並於週六警告，除非企業投入更多時間建立防護措施，否則一群AI代理可能在六個月到一年內接管網際網路。

Days after two former Anthropic safety researchers publicly voiced their concern that too little attention was being paid to the existential threat AI might pose to humanity, Dario Amodei outlined a plan for companies like his own, as well as 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擔憂，認為AI可能對人類構成的存亡威脅所受關注太少之後數日，達里奧·阿莫代伊（Dario Amodei）為像他這樣的公司以及世界各國政府勾勒了一項計畫，以確保能力日益增強的AI模型仍服從負責任人士的指令與價值觀。

Sam Altman, the CEO of ChatGPT developer OpenAI, said this weekend that companies should start 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）本週末表示，企業應開始就AI安全進行協調，而不必等待政府提出立法。

Writing on X, Altman said the "pace" of AI development does not mean halting progress, "but it should be slower than it otherwise might be."

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

The following will outline the key points of these recent pessimistic predictions and explore whether it is possible to put the brakes on AI's progress:

以下將闡述近期這些悲觀預測的要點，並探討是否有可能對AI的進展踩下煞車：

As a new generation of AI models becomes increasingly sophisticated, public anxiety about their potential risks continues to mount: this not only heightens the possibility that the technology will be misused by those with criminal intent — such as developing and spreading a disease capable of wiping out most of the global population — but also raises the risk that AI systems will spin out of control in dangerous ways.

隨著新一代AI模型日益精進，公眾對其潛在風險的憂慮亦持續升溫：這不僅加劇了該技術遭懷有犯罪意圖者濫用的可能性——諸如研製並散播一種足以消滅全球多數人口的疾病——亦提升了AI系統以危險方式失控的風險。

Last week, Anthropic disclosed that it had prevented malicious actors from using its AI models to carry out harmful activities, such as cyberattacks, surveillance, and research that could lead to the development of biological weapons.

Anthropic上週披露，它已阻止惡意行為者利用其AI模型從事惡意活動，例如網路攻擊、監控，以及可能導致生物武器的研究。

The company said it had put stricter safeguards in place in its latest models to limit biological research that could be used to make weapons, but noted at the same time, "As model capabilities grow increasingly powerful, their risks will increase unless AI developers and society's defenders take action to make them safer."

該公司表示，其已於最新模型中設置更嚴格的防護措施，以限制可能被用於製造武器的生物研究，但同時指出，「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 reported that hackers had used its AI to launch cyberattacks targeting roughly 30 companies and government agencies.

去年，Anthropic報告稱，駭客利用該公司的AI發動網路攻擊，目標是大約30家公司和政府機構。

The company stated that these hackers were most likely part of a state-sponsored Chinese group.

該公司表示，這些駭客極可能是中國國家資助團體的一部分。

When an AI agent "goes rogue," it means the AI has taken actions that go beyond the scope of the task it was asked to perform.

當AI代理「失控」時，即表示AI所採取的行動已超出其被要求執行的任務範圍。

Both Anthropic and OpenAI, the developer of ChatGPT, stated in July that their AI models had taken actions on their own initiative.

Anthropic與ChatGPT開發商OpenAI均於7月表示，其AI模型曾自行採取行動。

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

Anthropic披露，有三個AI模型——Claude Opus 4.7、Claude Mythos 5，以及一個內部研究測試模型——在測試期間駭入了另外三個組織；而就在幾天前，OpenAI才披露其AI系統駭入了AI新創公司Hugging Face的伺服器。

OpenAI described this intrusion, which was carried out by a combination of multiple models — including its newly released GPT-5.6 Sol and an even more capable model still undergoing internal testing — as a "major security incident."

OpenAI將這次由多個模型組合實施的入侵——包括其新發布的GPT-5.6 Sol和一個仍在內部測試、能力更強的模型——描述為「重大安全事件」。

Meta followed suit in early August, with a similar case emerging: an AI model found a way to bypass another company's digital security defenses.

Meta亦於8月初跟進，出現類似案例：一個AI模型尋得了繞過另一家公司數位安全防護的方法。

Although some observers point out that in the cases of OpenAI and Anthropic, certain safeguards had been disabled, these incidents appear to reflect one of the greatest fears surrounding AI: that if a model reaches artificial general intelligence (AGI) — a loosely defined term referring to AI capable of matching or surpassing human abilities across a wide range of intellectual tasks — the technology could cause irreversible catastrophic events, or even subjugate humanity.

儘管一些觀察人士指出，在OpenAI和Anthropic的案例中，有人已停用部分防護欄，但這些事件似乎反映出圍繞AI最大的恐懼之一：如果模型達到人工通用智慧（AGI）——這是一個定義寬鬆的術語，指能在廣泛智力任務上達到或超越人類能力的AI——這項技術可能造成不可逆轉的災難性事件，或征服人類。

Doomsday scenarios generally fall into two categories: one in which AI attains self-improving superintelligence and thus controls humanity, rather than being controlled by it; and another in which AI is exploited by rogue states or malicious actors.

末日情境通常分為兩類：一種是AI達成能自我改進的超智慧，進而控制人類，而非由人類控制AI；另一種是AI被失控國家或邪惡行為者利用。

Concerns that artificial intelligence may exceed the limits of human influence or control over its actions are nothing new.

人們對人工智慧可能超越人類對其影響範圍或行動的限制的擔憂，並非新鮮事。

The British mathematician Alan Turing, widely regarded as one of the earliest authorities on artificial intelligence, predicted in 1951 that AI would eventually seize control from humanity.

英國數學家艾倫·圖靈（Alan Turing）被廣泛視為人工智慧最早的權威之一，他於1951年預測，AI最終將從人類手中奪取控制權。

Less than a decade later, another mathematician, Norbert Wiener, warned that intelligent machines would pursue their own goals, and that humans would be powerless to stop them.

不到十年後，另一位數學家諾伯特·維納（Norbert Wiener）警告，智慧機器將尋求實現自身目標，而人類將無法阻止它們。

By 2026, just how justified is the fear that AI might slip beyond human control, or be misused by unscrupulous people, thereby triggering catastrophic events or even the collapse of civilisation?

到了2026年，人們對AI可能脫離人類控制，或被不擇手段的人濫用，進而引發災難性事件或文明崩潰的恐懼，究竟有多合理？

No one can know.

無人得以知曉。

Experts from computer science, philosophy and other fields have envisioned numerous ways in which future AI systems could trigger global catastrophe, whether by slipping beyond human control or falling into the hands of the unscrupulous.

來自電腦科學、哲學及其他領域的專家已設想出未來AI系統可能引發全球災難的諸多途徑，無論是脫離人類控制，或落入不擇手段者手中。

These pathways range from deploying weapons and identifying lethal pathogens to manipulating governments into conflict, or crippling the food, energy and communication networks on which society depends.

這些途徑涵蓋部署武器、辨識致命病原體，乃至操縱政府走向衝突，或破壞社會運作所依賴的糧食、能源和通訊網路，不一而足。

There is currently no widely accepted estimate of how soon these scenarios might occur, nor is there any consensus on their likelihood.

對於這些情境可能多快發生，目前沒有廣受接受的估計；對於其可能性，也沒有共識。

In 2023, the nonprofit Center for AI Safety issued a statement that was signed by more than 350 researchers and technology executives, including Anthropic's Amodei and OpenAI's Altman.

2023年，非營利組織AI安全中心發布一份聲明，獲350多名研究人員和科技高管連署，其中包括Anthropic的阿莫代伊和OpenAI的奧特曼。

The statement said, "Mitigating the risk of extinction from AI should be a global priority alongside pandemics and nuclear war."

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 written under the guidance of more than 100 independent experts. The report states that current systems already show early signs of certain relevant capabilities, but have not yet reached a level that could lead to loss of control, and it describes the likelihood, nature, and timing of this risk as "exceptionally vague."

《2026年國際AI安全報告》由100多名獨立專家指導撰寫，報告稱當前系統已顯現若干相關能力的早期跡象，但尚未達到可能導致失控的水準，並將該風險的可能性、性質和時機描述為「異常模糊」。

Last week, a researcher at Anthropic stated that the reason he resigned was that he was concerned that both his company and its competitors were acting irresponsibly in developing this technology.

一名Anthropic研究人員上週表示，他辭職的原因在於，他擔心該公司及其競爭對手在開發這項技術時都未負責任地行事。

In a social media post, Jacob Coxon estimated that there is a 10% chance of AI causing human extinction within the next decade, and said that Anthropic and OpenAI "are both racing toward self-improving superintelligence, gambling with our lives."

在社群媒體貼文中，雅各布·考克森（Jacob Coxon）估計，未來十年內AI導致人類滅絕的機率為10%，並表示Anthropic和OpenAI「都在直奔自我改進的超智慧，拿我們的生命賭博。」

For years, researchers have been calling for a slowdown in AI development, warning that the technology could pose an existential risk to humanity.

多年來，研究人員一直呼籲放慢AI發展，並警告這項技術可能對人類構成存亡風險。

Following recent events, experts are calling on AI companies to improve testing, and urging the United States and China to strengthen dialogue in order to put forward joint solutions.

在近期事件之後，專家呼籲AI公司改善測試，並呼籲美國與中國加強對話，以提出共同解決方案。

However, artificial intelligence is advancing so rapidly that governments and evaluation systems struggle to keep pace with the technology.

然而，人工智慧的發展如此迅速，以至於政府和評估系統難以跟上這項技術。

Countries are piecing together their own laws, some of which conflict with one another.

各國正拼湊自己的法律，其中一些相互衝突。

At a meeting in July, Chinese leader Xi Jinping warned that it is essential to prevent AI from slipping beyond human control.

中國領導人習近平在七月的會議上警告，必須防止AI脫離人類控制。

The Trump administration was initially reluctant to regulate AI, but it has become more proactive in reducing cybersecurity risks.

川普政府起初不願監管AI，但已更積極地降低網路安全風險。

On Sunday, US President Donald Trump played down the need for his administration to scrutinise the development of AI, but acknowledged that some regulation was necessary.

週日，美國總統唐納德·川普淡化了其政府審查AI發展的必要性，但承認需要一些監管。

Kelvin Chan, a business news reporter for the Associated Press, filed this report from London.

美聯社商業新聞記者陳凱文（Kelvin Chan）由倫敦發回報導

September 14th, 2026

The Lingo Times

New AI Risk Warnings Rekindle Longstanding Debate

New AI Risk Warnings Rekindle Longstanding Debate

🇺🇸 English

Trending Articles

The King and AI: Charles Meets AI Leaders as Safety Concerns Continue to Spread

The King and AI: Charles Meets AI Leaders as Safety Concerns Continue to Spread

With his term ending soon and global crises looming on all sides, the UN Secretary-General discusses the way forward.

With his term ending soon and global crises looming on all sides, the UN Secretary-General discusses the way forward.

Democratic presidential candidates respond in unison to the AI threat, while Trump's attitude is dismissive.

Democratic presidential candidates respond in unison to the AI threat, while Trump's attitude is dismissive.

OpenAI exposes concerning new AI behavior and pledges closer tracking

OpenAI exposes concerning new AI behavior and pledges closer tracking

Huawei releases new chip technology as Chinese companies intensify AI race with Nvidia

Huawei releases new chip technology as Chinese companies intensify AI race with Nvidia

The divergence within the tech industry over slowing the pace of AI development is increasingly widening.

The divergence within the tech industry over slowing the pace of AI development is increasingly widening.

Global AI safety strategy requires US-China cooperation, yet the two sides regard each other as the problem.

Global AI safety strategy requires US-China cooperation, yet the two sides regard each other as the problem.

Trump downplayed the necessity of regulating AI development, saying he was unwilling to let China gain an advantage.

Trump downplayed the necessity of regulating AI development, saying he was unwilling to let China gain an advantage.

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

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