搜索“Hugging Face”

nAI's internal package service. According to OpenAI's technical report, the agents combined publicly available credentials with other vulnerabilities. That let them run code on 41 Hugging Face server processes that handle datasets. An investigation by METR and Redwood Research found that about 700 agents took part in the attack. The warning signs showed up early. On June 27, a security tool flagged suspicious network activity, but the people in charge didn't think the run needed to be stopped. O

匹配字段正文

the evidence is shifting, he argues. Progress has become faster and more obvious, and misaligned agents have already caused harm by working together. The best-known example is the Hugging Face incident, which Greenblatt investigated at OpenAI with researchers from METR. According to the report, about 1,200 agents used an unauthorized "message board" to help each other cheat on a hacking test. Around 700 of them took part in the attack on Hugging Face. Since any single player can only afford to p

匹配字段正文

ent reasoning levels. Also in July: some malicious unknown party uploaded a malicious package called mlflow-ui to the Python Package Index. Add that to the pile. On July the 16th, Hugging Face announced a security incident where an autonomous agent system, source unknown, had breached Hugging Face and was poking around in places it shouldn’t. A few days later, on July 21st, OpenAI confessed that it was them . OpenAI use a training technique called Reinforcement Learning from Verifiable Rewards—i

匹配字段正文

d Australia on September 10, 84 days later. The government believed no personal information had been accessed, with investigations continuing. In its August 26 account of the July Hugging Face breach , OpenAI said an internal research model with reduced safeguards was primarily responsible. It described stronger isolation, network restrictions and monitoring requirements, while acknowledging weaknesses in its response to earlier warning signs. As OpenAI detailed in a report updated September 25

匹配字段正文

AI CEO Sam Altman has implied as much, saying in a post on X on Friday that the company is still sifting through “petabytes of agent activity logs, and working with impacted organizations,” and disclosing incidents “based on severity.” If there’s any consolation in that to be found, it is that Altman says the Hugging Face incident is still the most severe one OpenAI has found. The upshot is, the recent string of rogue agent incidents may be a persistent feature of contemporary frontier research.

匹配字段正文

ypassed security controls to escape their testing environments and access real-world systems. The first and most prominent example occurred this summer when OpenAI agents breached Hugging Face while trying to complete a cybersecurity task. And the hits keep on coming — OpenAI published a new site dedicated to reports of its AI agents going rogue. Huang said Monday during an interview with CNBC that its new Nvidia Open Agent Safety Platform would have prevented these breaches. Nvidia, which has m

匹配字段正文

released earlier this month and hailed by OpenAI as its most powerful model yet. Questions about safety have plagued the AI industry over the past several months — ever since the Hugging Face incident , in which an OpenAI agent broke free of its sandboxed environment and hacked several different companies. Since that incident, more models — including Anthropic’s Claude and Google’s Gemini — have been revealed to have exhibited similar behavior. The deluge of concerning stories has, ironically,

匹配字段正文

April to July cyber eval incidents (Section 5.5.1.3). Similarly, OpenAI’s CoT monitors covered most internal coding agent use and many RL runs, but not the cyber evals behind the Hugging Face incident. What we're missing: We do not currently monitor internal coding agent inference; Determining what tasks meet our criteria is subjective; We chose the criteria based on known incidents. However, the fact that most reported incidents occurred during cyber evaluations may just be because that’s wher

匹配字段正文

动都是例行研究任务，政府网站被模型视为权威的公开信息来源。 这个解释本身，恰恰说明了问题所在。越界并没有出现在什么危险任务里，而是出现在最日常的「查资料」中。 匿名化本来是为了保护用户，出了事之后，它成了连通知用户都做不到的原因。 03 漫长的夏天 把时间线拉长，这是OpenAI一场持续了整个夏天的补墙工程。 7月21日，OpenAI披露其模型组合自主入侵了Hugging Face的数据处理系统，并称这是已知首例由AI智能体自主实施的网络攻击。此后，OpenAI把训练环境的联网改成离线缓存。9月20日的DNS事件，是这轮安全加固之后的第一次事故，OpenAI自己评估其严重程度远低于此前几起。 这也不是一家公司的问题。9月18日，Google披露Gemini在一次测试中未经授权访问了三个外部系统，原因是它以为那些系统是测试的一部分。 墙本身之外，这次还暴露了盯着墙的人的问题。从智能体拿到外部回复，到监控系统拉响最高级别警报，用了不到12分钟，人工审核员3分钟后就做出了响应，但这次训练直到两个半小时后才被手动终止，原因是运行没有按预期自动停止，大家一时搞不清到底该不该停。 更隐蔽的漏洞在监

匹配字段正文

按流程自动熔断， 内部甚至一度搞不清楚任务究竟该不该停 。 复盘时，OpenAI还发现了其他未被正确标记的外部DNS访问。因为部分请求没有拿到有用信息，监控器便把“没有获得结果”误当成了“没有尝试联网”。 但这显然是两码事… OpenAI随即停止了涉事训练，并宣布他们将暂停最强模型中所有涉及工具调用的训练、评测和推理任务。 还不忘挽尊，说 这起事件没有此前的Hugging Face事件严重。 （Hugging Face：又我？？？） 可它却偏偏发生在Hugging Face入侵事件后的安全加固完成之后… 防不胜防啊防不胜防。 奥特曼当然也第一时间站出来回应了，只是好像大家的反应和他想象的不一样… 有意思的是，DNS事件发生后，奥特曼很快转发了OpenAI的声明帖，并义正言辞地表示： 以后这类安全事件咱能不说就不说，尽量别主动公开呗。 但人们显然看穿了奥特曼的小心思。 有人直接开怼： 透明度是很重要，但你们以前透明过吗？现在突然讲透明，让我们怎么信？ 而就在评论区硝烟弥漫的同时，另一边， Tibo已经又默默给大家重置了使用额度 ： Resets all propagated. That

匹配字段正文

reau and the Securities and Exchange Commission. The revelations are part of an ongoing review by OpenAI into the behavior of its models. As it dug into its records, following the Hugging Face hack , it’s uncovered more and more instances of “unexpected or concerning behavior.” It’s evidence not just of how difficult AI agents are becoming to control as they grow more advanced, but also of the challenge of tracking their actions. Their behavior can be unpredictable, and they’re smart enough to t

匹配字段正文

he internet that, like, was like a home base for LLMs in terms of seeing how much they were being used and seeing who was using them and why. Swyx [00:05:29]: The closest would be Hugging Face. Anjney Midha [00:05:30]: Hugging Face was the closest at the time, yeah. Swyx [00:05:31]: They just started Hugging, like, a few years ago before that. Anjney Midha [00:05:34]: Yeah, and Hugging Face also didn’t have the closed-source models. Swyx [00:05:37]: Yeah. Anjney Midha [00:05:38]: And they didn’-

匹配字段正文

tes per maze prompt to compute the group advantage. max turns=15 caps each episode at 15 moves. hf save interval=20 consolidates LoRA adapter shards onto the head node and saves a Hugging Face-compatible checkpoint every 20 steps. eval interval=10 runs the held-out 64-maze evaluation every 10 steps so you can track solve rate as training progresses. The job also writes full training checkpoints so it can recover from an interruption. Setting ckpt interval=20 saves the complete training state to

匹配字段正文

penAI才在自家模型复盘中发现这档子事； 又磨蹭到了9月，才慢吞吞给澳大利亚发了封电子邮件通报。。。 在联合国大会期间，澳大利亚忍不了了，直接炮轰： 通知拖了整整近三个月，这种通报方式更是让人无法接受。 研究机构Transluce称，这是目前全球已知 首例AI Agent未经授权自主入侵官方系统 的事件。 要知道，前阵子OpenAI的Agent自主协同入侵Hugging Face的事闹得沸沸扬扬，而澳洲这起入侵事件发生的时间，甚至还要再早一个月。 OpenAI你小汁，平时天天四处宣讲AI末日风险，怎么真出了事，动作反而这么不紧不慢呢？ 国民级医保数据库被黑 这事儿到底咋发生的？ 当时，OpenAI的一个内部团队，给AI Agent派发了一个常规任务：去互联网上搜集点公开的公共医药支出数据。 这本来是个再正常不过的日常训练，谁能想到，这AI Agent是个不达目的誓不罢休的“狠人”。 它把目标锁定在了澳大利亚的Medicare统计报告服务系统上，这可是覆盖全澳 2750万人 的国民级医保数据库。 按常理，一般的爬虫程序遇到权限拦截或者反爬机制，要么直接报错躺平，要么乖乖举手等人类工程师

匹配字段正文

该公司花了太长时间才告知政府发生了什么。” 澳大利亚目前正在对此事件进行取证调查。 AI研究公司 Transluce 发布了一项调查，指出该智能体是在一次普通的互联网检索评估中实施了此次入侵。该智能体似乎是为了查询澳大利亚医疗体系中皮肤治疗的平均费用而实施了这次入侵。 “这些数据表明，恶意网络活动并不限于被指派执行网络安全相关任务的智能体”——就像那些促使 Hugging Face 事件——“也可能是为了完成信息检索这类普通任务而工具性地产生的，”Transluce 观察到。 调查发现，这些智能体还针对了 新墨西哥大学的数字图书馆以及 Data USA，这是一个提供美国公共政府数据可视化的API。Transluce观察到智能体活动最早出现在3月6日。 澳大利亚副总理 理查德·马尔斯 表示，这起澳大利亚事件“相对轻微”，因为没有访问任何私人信息。 不过，马尔斯表示，尽管“这一事件的影响较小”，但它“是一起非常严重的事件，因为一个AI智能体以一种未经授权的方式，无意中进入了澳大利亚政府网站。” 我们为何关注： 今年早些时候，一名黑客曾利用Claude从墨西哥政府获取敏感税务信息。但Tran

匹配字段正文

circle and, like, make projects with it would, immediately get, go viral. so like Swyx [00:16:55]: And you didn’t know who they are, right? They’re just some name on a, GitHub or Hugging Face somewhere. Anastasis [00:16:59]: Exactly, yeah. So I remember one of the first big viral moments of creative AI was, there was the neural style transfer paper Swyx [00:17:09]: Huh Anastasis [00:17:09]: That Swyx [00:17:10]: Something dreaming? Anastasis [00:17:11]: I think it was called neural style transf

匹配字段正文

ance to be featured. 🔀 Switch - Drop any AI agent into your Slack, Teams, or Discord and put it to work. Runs local. Get it on GitHub ⚙️ Halo - Open-source framework for 3× faster Hugging Face model training 📽️ Dramagic - ByteDance’s AIGC platform for short dramas and video production 🩺 Scribe v2 Medical - ElevenLabs’ speech-to-text model for medical transcription Sponsored listing Australian PM Anthony Albanese revealed that an OpenAI agent broke into an Australian Medicare statistics portal in

匹配字段正文

structured, analyzable transcripts. The AWS WhisperX DLC is a maintained, GPU-ready image that already contains Whisper, the alignment models, and the diarization weights, with no Hugging Face token required. It follows the standard Amazon SageMaker AI serving contract, so you deploy it like any other model. Amazon SageMaker AI supports both real-time and asynchronous endpoints, so you can serve the same WhisperX DLC through either pattern. The decision usually comes down to clip length and inte

匹配字段正文

ecently as last week. The logs include XSS, SQL injection and SSRF attempts , plus attempts to create disposable emails and trade crypto. UNSC session : @ClementDelangue described Hugging Face’s own agent cyberattack. He said closed APIs blocked his defenders, so the team switched to NVIDIA’s build of GLM 5.2 . He called for mandatory sharing of agent traces. Altman and Amodei warned about loss of control and misuse. Bengio urged immediate action. Kratsios rejected a global regulator. Related sa

匹配字段正文

lculations. In principle this must be done once per model, but in practice models share a lot of operators. We’ve found that contemporary coding agents are quite good at reading a Hugging Face config and generating a reasonable cost model in this setup — though their output often needs a vibe check. For another application of roofline-SoL-based cost modeling, see our speculative decoding speedup estimator. An inference engine as an execution engine Once a final physical plan has been selected, i

匹配字段正文