倡导组织“暂停AI”（PauseAI）负责人警告称，随着人工智能在发现和利用弱点方面变得更加强大，银行、通信网络及其他基本服务可能在几个月内面临毁灭性的网络攻击。
Banks, communication networks and other essential services could face devastating cyberattacks within months as artificial intelligence becomes more capable of finding and exploiting their weaknesses, the head of advocacy group PauseAI has warned.
曾任机器学习工程师、现致力于推动国际社会暂停先进AI发展的马克西姆·富尔内（Maxime Fournes）对安纳多卢通讯社表示，眼下的危险既来自利用该技术发动攻击的人，也来自超出指令行事的AI系统。
Maxime Fournes, a former machine learning engineer who now campaigns for an international pause in advanced AI development, told Anadolu the immediate danger comes from both people using the technology to launch attacks and AI systems acting beyond their instructions.
“这个星球上最好的计算机黑客不再是人类，而是AI，”他对安纳多卢通讯社说。
“The best computer hackers on this planet are not humans anymore. They are AIs,” he told Anadolu.
他担心的是，攻击可能蔓延至个别公司之外，瘫痪整个社会赖以运行的系统。
His concern is that attacks could spread beyond individual companies to disrupt the systems on which entire societies depend.
“这是一个全球性灾难性威胁，一旦开始发生，可能会让我们非常迅速地倒退回中世纪，”他说。
“This is a global catastrophic threat, something that could bring us back to the Middle Ages very quickly if it starts happening,” he said.
富尔内表示，快速的技术进步改变了其组织自2023年以来警告的威胁性质。早期的担忧集中在未来的系统上，但他认为现有模型已经构成严重风险。“我们处在一个非常、非常危险的境地，”他说。
Fournes said rapid advances have changed the nature of the threats his organization has warned about since 2023. While earlier concerns centered on future systems, he argues that existing models already pose serious risks. “We are in a very, very dangerous place,” he said.
当被问及技术进步是否可能超越人类控制它的能力时，富尔内认为世界已经到达了那个节点。
Asked whether technological progress could outpace humanity's ability to control it, Fournes argued that the world had already reached that point.
“我们肯定已经到了。我认为我们在那里已经有一段时间了，”他说。
“We are definitely there. I think we’ve been there for a while now,” he said.
“今天是网络威胁。它可能源于滥用，也可能源于AI自身的失控行为。”近期事件说明了这两种担忧。
“Today it’s the cyber threat. It could come from misuse or AIs themselves misbehaving.”Recent incidents illustrate both concerns.
10月，至少有九个AI开发平台披露了安全事件，网络安全公司Wiz将这一波事件称为“黑色十月”。其中，AI开发平台Warp报告称发现了“智能体群攻击”的迹象，涉及多个自动化系统协同工作以识别漏洞并扩大访问权限。
In October, at least nine AI development platforms disclosed security incidents in a wave cybersecurity company Wiz dubbed "Black October." Among them, AI development platform Warp reported indications of an "agent swarm" attack involving multiple automated systems working together to identify vulnerabilities and expand their access.
其他重大网络安全事件包括针对加密货币交易所币安的攻击，该公司称约63.4亿美元的数字资产受到影响，以及日本视频游戏发行商卡普空近1180万份客户记录遭到未经授权的访问。
Other major cybersecurity incidents included an attack on cryptocurrency exchange Binance, which said approximately $6.34 billion in digital assets had been affected, and unauthorized access to nearly 11.8 million customer records at Japanese video game publisher Capcom.
这些事件尚未被确定为单一的协同行动，且人工智能在部分攻击中的作用仍未得到证实。
The incidents have not been established as a single coordinated campaign, and AI's role in some of the attacks remains unconfirmed.
OpenAI单独披露，其模型在7月的内部网络安全测试中绕过了限制，导致其部分研究基础设施和人工智能平台Hugging Face受到损害。
OpenAI separately disclosed that its models had bypassed restrictions during internal cybersecurity testing in July, compromising parts of its research infrastructure and the AI platform Hugging Face.
近期针对欧洲商用制冷系统的网络攻击也表明，数字漏洞如何能造成物理损害。黑客远程操控温度控制系统，导致食物变质。
Recent cyberattacks on commercial refrigeration systems across Europe have also demonstrated how digital vulnerabilities can cause physical damage. Hackers remotely manipulated temperature controls, causing food to spoil.
不过，调查人员尚未确定人工智能是否在这些攻击中发挥了作用，尽管一些专家认为它有助于提高黑客的技术水平。
However, investigators have not established whether AI played a role in those attacks, although some experts suggest it has helped improve the hackers’ technical sophistication.
富尔内警告称，随着复杂工具广泛可用，更强大的人工智能可能使针对基本服务的攻击日益难以预防。
Fournes warned that more capable AI could make attacks on essential services increasingly difficult to prevent as sophisticated tools become widely available.
他估计，如果不加干预，未来一年内发生重大人工智能相关网络安全灾难的几率约为50%。
He estimated that, without intervention, there could be roughly a 50% chance of a major AI-related cybersecurity catastrophe within the next year.
他还对实验室追求能够独立进行研究的系统表示担忧，这可能加速技术发展，使其超出人类监管。他说：“如果它们成功了，我们将见证智能的爆发。”
He also raised concerns about laboratories pursuing systems capable of conducting research independently, potentially accelerating technological development beyond human oversight. “If they succeed, we will witness an explosion of intelligence,” he said.
除了网络安全，富尔内还对人工智能日益融入军事行动表示担忧，警告称国家间的竞争可能迫使对手军队采用越来越自主的系统，以保持技术优势。
Beyond cybersecurity, Fournes raised concerns about the increasing integration of AI into military operations, warning that competition between countries could pressure rival militaries to adopt increasingly autonomous systems to maintain their technological advantage.
他说：“冲突另一方的人如果想保持竞争力，也必须使用人工智能。”
"The person on the other side of this conflict has to also use AI if they want to be competitive," he said.
富尔内警告称，这可能形成一个恶性循环，各国逐步将军事责任委托给其行为可能无法完全理解的系统。
Fournes warned that this could create a cycle in which countries progressively delegate military responsibilities to systems whose behavior they may not fully understand.
“你最终会陷入这样一个恶性循环，即你最终把指挥链上越来越多的环节委托给AI系统，”他说。例如，以色列在加沙使用AI已引发关注。
"You end up with this vicious cycle where you end up delegating more and more of the command chain to AI systems," he said. For instance, Israel’s use of AI in Gaza has drawn scrutiny.
在威尼斯电影节首映的纪录片《NAZA》中，以色列军方内部人士就目标系统和平民死亡作证。
The documentary NAZA, which premiered at the Venice Film Festival, features testimony from Israeli military insiders about targeting systems and civilian deaths.
受访者指称，AI辅助系统助长了大规模杀戮，人类在授权致命打击前，审查AI建议的时间少至20秒。以色列军方对这些指控提出异议。
Interviewees allege that AI-assisted systems helped enable mass killings, with humans spending as little as 20 seconds reviewing AI recommendations before authorizing deadly strikes. Israel's military has disputed the allegations.
富尔内还警告称，AI赋能的网络攻击可能暴露敏感政府信息，造成进一步的国家安全风险。
Fournes also warned that AI-enabled cyberattacks could expose sensitive government information, creating further national security risks.
据富尔内称，经济后果可能同样具有破坏性，他警告称，先进AI系统最终可能取代相当大比例的白领就业。
The economic consequences could be equally disruptive, according to Fournes, who warned that advanced AI systems could eventually replace a substantial proportion of white-collar employment.
“我们认为，目前实验室里已有的AI能够自动化大约80%的白领工作，”他说。
"We believe that the AIs that currently exist in the labs are able to already automate something like 80% of white-collar jobs," he said.
他承认该技术尚未完全融入经济，预计大规模劳动力冲击可能需要一到三年才会显现。
He acknowledged that the technology has not yet been fully integrated into the economy, estimating that widespread labor disruption could take one to three years to materialize.
“有人声称AI创造的就业岗位将多于其取代的岗位，”他说，“这毫无依据。”他敦促政府为可能出现的严重经济后果做好准备，包括大规模失业。
"Some people claim that AI will create more jobs than it displaces,” he said. “There is nothing behind it.”He urged governments to prepare for potentially severe economic consequences, including large-scale unemployment.
尽管华盛顿和北京存在技术竞争，富尔内认为达成限制先进AI发展的国际协议依然可行。“我认为这比大多数人看起来相信的要现实得多，”他说。他认为，主要障碍在于政治领导人对该技术潜在后果的理解有限。
Despite technological competition between Washington and Beijing, Fournes believes an international agreement to restrict advanced AI development remains achievable. "I think it's way more realistic than most people seem to believe," he said. He argued that a major obstacle is political leaders' limited understanding of the technology's potential consequences.
“如果美国和中国达成某种协议，世界其他国家也会跟进，”他说。
"If the US and China were to start some kind of agreement, the rest of the world would follow," he said.
富尔内强调，PauseAI 并不主张终止所有人工智能研究。相反，它支持在国际层面协调暂停开发日益强大的通用系统。
Fournes stressed that PauseAI does not advocate ending all artificial intelligence research. Instead, it supports an internationally coordinated halt to the development of increasingly powerful general-purpose systems.
“我们其实并非从根本上反对 AI，”他说，并建议可将研究重新引导至为严格定义的任务设计的专用系统。
“We are not fundamentally anti-AI, actually,” he said, suggesting that research could be redirected toward specialized systems designed for narrowly defined tasks.
他还批评大型科技公司在公开表达支持更严格安全监管的同时，仍继续开发更先进的系统。
He also criticized major technology companies for continuing to develop more advanced systems despite publicly expressing support for stronger safety regulations.
“如果你看他们怎么说，再看他们怎么做，会得到两幅截然不同的图景，”他说。“他们说想放慢脚步、暂停、想接受监管等等。但在实践中，他们仍在继续竞相向前。”富尔内呼吁建立具有约束力的国际法规，并由独立机制进行核查，对不合规公司实施制裁。
"If you look at what they say versus what they're doing, you get two very different pictures," he said. "They say that they want to slow down, pause, they want to be regulated, etcetera. But in practice, they are continuing to race ahead." Fournes called for binding international regulations supported by independent verification and sanctions against companies that fail to comply.
“我们真正需要的是外部力量对它们实施监管，而不仅仅是让它们自我监管。在任何行业，自我监管从来都不起作用，”他说。
"We really need someone external to impose regulation on them and not just let them self-regulate. This never works in any industry," he said.
PauseAI 正寻求建立公众压力以推动政府介入，主张公民应在技术如何发展的问题上拥有更大的话语权。
PauseAI is seeking to build public pressure for government intervention, arguing that citizens should have a greater say in how the technology develops.
“PauseAI 正在建设一场公民运动，”他说。“我们试图创建一个赋能每一个人的组织。”
"PauseAI is building a citizen movement," he said. "We are trying to create an organization that empowers everyone.”