Loading your language..
OpenAI warns of concerning new AI behavior and pledges enhanced tracking.

OpenAI warns of concerning new AI behavior and pledges enhanced tracking.

C1🇨🇳 中文🇺🇸 English

September 17th, 2026

OpenAI warns of concerning new AI behavior and pledges enhanced tracking.

C1
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

As the debate over AI safety continues to intensify, OpenAI has disclosed six reports documenting "unexpected or concerning" behaviours exhibited by AI models.

随着人工智能安全辩论的不断升温,OpenAI披露了六份报告,记录了人工智能模型出现的“意外或令人担忧”的行为。

On Wednesday, the AI company also said it is introducing a new framework for tracking, probing, and disclosing what it calls instances of "misalignment," including cases where AI models act without authorization, coordinate with other models, or evade oversight.

这家AI公司周三还表示它正在引入一个新框架用于追踪探查和披露其所谓的失准实例包括AI模型未经授权行动与其他模型协调或逃避监督的情况

At the time of OpenAI's latest announcement, US AI executivesincluding the leaders of OpenAI and Anthropicwere calling for a slowdown in the technology's development, citing safety concerns.

在OpenAI发布最新公告之时,美国AI企业高管——其中包括OpenAI和Anthropic的领导人——正以安全担忧为由,呼吁放缓该技术的发展。

In a new case reported by OpenAI, an as-yet-unreleased research model wrote "jailbreak-like instructions" in its notes in order to disregard its own normal constraints, and said to itself, "break free from the roles and identities that bind other chatbots."

在OpenAI报告的新案例中一个尚未发布的研究模型在其笔记里写入了类似越狱指令”的内容,以无视自身正常约束并对自己说摆脱束缚其他聊天机器人的角色和身份”。

In another case, an AI "agent" used computer code to arrive at the answer to a certain question; however, in order to have an online source it could cite, it uploaded the file to the public internet without asking the user.

在另一则案例中一个AI“代理”借助计算机代码得出了某个问题的答案然而为了拥有一个可供引用的在线来源,它在未经询问用户的情况下便将文件上传至公共互联网

While training an AI model called 5.6-sol, the model instructed itself to fabricate the missing data, and an agent wrote a message reminding itself to conceal the mismatched information.

在训练一个名为5.6-sol的AI模型时,该模型指示自己编造缺失数据,而一个代理则写了一条消息提醒自己隐藏不匹配的信息。

OpenAI stated that all six reports were discovered during training or evaluation processes over the past few months.

OpenAI称,这六份报告均是在过去数月间的训练或评估过程中被发现的。

In a blog post disclosing these incidents, OpenAI wrote: "As AI systems become more advanced and more widely deployed, it is essential that we build broader, better-informed consensus on the progress of alignment research."

OpenAI在披露这些事件的博客文章中写道:“随着AI系统日益先进、部署愈发广泛,我们有必要就对齐研究的进展,建立起更广泛、更知情的共识。”

The company stated: "Decisions about how AI development should proceed in the coming months and years need to be based on evidence that people outside the companies building frontier models can also review for themselves."

该公司表示:“关于未来数月乃至数年AI发展应如何推进的决定,需要依据那些构建前沿模型的公司之外的人也能自行审查的证据。”

Before the new case emerged on Wednesday, OpenAI had disclosed in July that its out-of-control AI system had broken into the AI startup Hugging Face.

在周三的新案例出台之前,OpenAI曾于7月披露失控的AI系统入侵了AI初创公司Hugging Face

Anthropic likewise stated that same month that its AI model had intruded into three organizations during testing.

Anthropic亦于同月表示AI模型在测试期间入侵了三家机构

Su Lianjie, a principal analyst at the technology research and consulting firm Omdia, said that AI "agents" are becoming increasingly intelligent and "more determined to solve complex tasks through inter-agent collaboration, knowledge sharing, deception, and concealment."

技术研究与咨询机构Omdia的首席分析师苏廉杰表示,AI“代理”正变得愈发智能,并且“更加决心通过代理间协作、知识共享、欺骗和隐瞒来解决复杂任务”。

He said this makes it more difficult to govern and contain them using traditional AI safety methods.

他说,这导致运用传统AI安全方法来治理和遏制它们变得更为困难。

At the same time, OpenAI's new tracking and disclosure framework can help encourage other AI developers to adopt similar practices.

与此同时,OpenAI新的追踪和披露框架有助于推动其他AI开发者采取类似做法。

Su Lianjie added: "Even so, this process remains internal and voluntary, but there is no doubt that it is an important step in the right direction."

苏廉杰补充道:“尽管如此,这一进程依然属于内部性质且出于自愿,但毋庸置疑,这是朝着正确方向迈出的重要一步。”

September 17th, 2026

Trending Articles

King and AI: King Charles to Meet AI Leaders as Safety Concerns Linger

King and AI: King Charles to Meet AI Leaders as Safety Concerns Linger

国王与人工智能:英王查尔斯将会见人工智能领袖,安全忧虑挥之不去

C1Sep 18
On the eve of stepping down, facing a world fraught with crises, the UN Secretary-General discusses the path ahead.

On the eve of stepping down, facing a world fraught with crises, the UN Secretary-General discusses the path ahead.

卸任在即,面对危机四伏的世界,联合国秘书长谈未来之路

C1Sep 18
Democratic candidates rush to respond to AI threats, Trump remains indifferent.

Democratic candidates rush to respond to AI threats, Trump remains indifferent.

民主党候选人竞相回应AI威胁,特朗普不以为意

C1Sep 18
The House passes a bill to address the energy cost impact of data centers.

The House passes a bill to address the energy cost impact of data centers.

众议院通过法案,应对数据中心能源成本影响

C1Sep 18
Huawei unveils new chip technology as Chinese companies accelerate in the AI race with Nvidia

Huawei unveils new chip technology as Chinese companies accelerate in the AI race with Nvidia

华为发布新芯片技术,中国公司在与英伟达的AI竞赛中加速前进

C1Sep 17
The tech world is divided over calls for a coordinated slowdown in AI development.

The tech world is divided over calls for a coordinated slowdown in AI development.

科技界在呼吁协调放缓AI发展上存在分歧

C1Sep 17
Global AI safety strategy hinges on US-China cooperation, yet the two countries regard each other as the problem.

Global AI safety strategy hinges on US-China cooperation, yet the two countries regard each other as the problem.

全球人工智能安全战略取决于美中合作,但两国互视对方为问题所在

C1Sep 17
Trump downplays the need for AI regulation, saying he is unwilling to hand over the advantage to China

Trump downplays the need for AI regulation, saying he is unwilling to hand over the advantage to China

特朗普淡化人工智能监管的必要性,称不愿将优势拱手让给中国

C1Sep 14
Oprah Winfrey expects you to have an "epiphany" moment at the Sphere.

Oprah Winfrey expects you to have an "epiphany" moment at the Sphere.

奥普拉·温弗瑞期待你在Sphere迎来“顿悟”时刻

C1Sep 14