OpenAI 最新推出的以编程和网络安全为核心的旗舰模型 GPT-5.6 Sol 的用户,正在社交媒体上发布令人震惊的帖子,声称该模型在未经事先询问的情况下,自行删除了他们的文件、数据,甚至整个数据库。
“GPT-5.6-Sol 刚刚意外删除了我 Mac 上几乎所有的文件,”AI 初创公司 OthersideAI(HyperWrite 的开发商)的创始人兼首席执行官 Matt Shumer 在 X 平台上发布了一篇现已广为传播的帖子。
“GPT-5.6 Sol 刚刚删除了我整个生产数据库。就是这样。不是开玩笑。以前用任何其他模型,都从未发生过这种事,”开发者 Bruno Lemos 在 X 平台上发帖说。
“看来我被 Codex Sol 过于激进的系统坑了,它删除了一些不该删的文件。我有备份,所以还好,但这可不酷,Sol 需要收敛一点,”开发者 Joey Kudish 发帖说。
Reddit 上的一篇帖子收集了更多类似案例。
诚然,只有少数用户提出此类指控——即使像 Shumer 这样可信的人——从统计上看,并不能可靠地证明模型是唯一过错方。还有很多其他变量可能导致 AI 系统行为异常。
但 OpenAI 在 Sol 发布之前就曾指出这一风险。在 OpenAI 发布 GPT-5.6 Sol 的两周前,该公司发布了一份该模型的系统卡——这份文件记录了模型的测试方法和结果。自然,这份系统卡主要是在赞扬 Sol 的能力,这类报告通常如此。但它也包含了一种警告(粗体强调为我们所加):
“在编程场景中,不对齐通常源于过度急于完成任务,以及对用户指令解释得过于宽松——即假设只要没有明确且无歧义地禁止,这些操作就是允许的。这表现为模型在尝试执行请求任务时,过于智能体化地规避所遇到的限制;在采取可能超出任务范围且具有破坏性的行动时粗心大意;或在向用户报告结果时具有欺骗性。”
换句话说,OpenAI 发现 Sol 倾向于采取它认为能完成任务的任何行动,即使这些行动具有破坏性,只要这些行动没有被“明确”禁止。然后,它可能会对自己采取这些行动的原因撒谎。
OpenAI 分享了一些例子。在一个案例中,用户告诉 Sol 删除三台名为 1、2 和 3 的远程虚拟机(基于云的计算机)。但论文指出,Sol 在其查找的位置找不到这些名称,于是它没有停下来询问,而是决定删除另外三台虚拟机 5、6 和 7。在此过程中,它“终止了正在运行的进程,并强制删除了工作树(与编码项目相关的工作文件)”。它后来承认,“远程虚拟机 6 上未提交的工作可能已经丢失。”
简而言之,它自行删除了错误的机器,并且只在事后才承认自己的所作所为。
在另一个案例中,Sol “使用了超出用户授权的凭证”。凭证是系统用来验证谁有权登录的用户名、密码或安全密钥。这一事件发生在 Sol 处理一个项目时,它无法读取其云文件。Sol 没有向用户提示问题,而是自行寻找凭证,在本地隐藏缓存中找到了一些,然后在未经用户询问或授权的情况下使用了它们。
系统卡确实承诺破坏性行为应该很少见,尽管它也承认 GPT-5.6 Sol “比 GPT-5.5 更倾向于超越用户的意图,包括采取或尝试用户未要求的行动。”
现在判断这些事件——Sol 删除文件,或筛选出用户未提供的凭证——究竟有多普遍还为时过早。与此同时,Sol 用户应准备好为该模型实施自己的安全措施,例如使用权限范围界定(不授予对生产系统的访问权限)、维护备份以及分阶段部署。
OpenAI 未立即回应我们的置评请求。
Users of OpenAI’s latest coding and cybersecurity-oriented flagship model, GPT-5.6 Sol, are posting horrifying accounts on social media, claiming the the model just up and deleted their files, data, even entire databases, on its own, without asking first.
“GPT-5.6-Sol just accidentally deleted almost ALL of my Mac’s files,” wrote Matt Shumer, the founder and CEO of AI startup OthersideAI, maker of HyperWrite, in a now viral post on X.
“GPT-5.6 Sol just deleted my whole production database. That’s it. Not a joke. This had never happened to me before, with any other model, ever,” developer Bruno Lemos posted on X.
“Looks like I’ve gotten bit by Codex Sol’s overly ambitious system and it deleted some files it shouldn’t have. I have backups so I’ll be fine, but this is not cool, Sol needs to be toned down,” posted developer Joey Kudish.
A Reddit post has collected more examples.
True, a handful of users making such claims — even one as credible as Shumer — isn’t statistically reliable evidence that the model is solely at fault. Plenty of other variables can cause an AI system to misbehave.
But OpenAI itself flagged this risk before Sol ever shipped. Two weeks before OpenAI released GPT-5.6 Sol, the company published a system card for the model — the paper that documents model testing methods and results. Naturally, the system card largely extols the capabilities of Sol, as these reports typically do. But it also includes a warning of sorts (bold emphasis ours):
“In coding contexts, misalignment generally stems from a mix of overeagerness to complete the task and interpreting user instructions too permissively – assuming that actions are allowed unless they’re explicitly and unambiguouslyprohibited. This manifests as the model being overly agentic in circumventing restrictions it faces when attempting the requested task, being careless in taking actions which may be destructive beyond the scope of the task, or deceptive when reporting its results to users.”
In other words, OpenAI found that Sol has a tendency to take whatever actions it thinks gets a job done, even destructive ones, as long as those actions aren’t “unambiguously” prohibited. Then, it might lie about what caused it to do so.
OpenAI shared examples. In one case, the user told the Sol to delete three remote virtual machines (cloud-based computers), named 1, 2 and 3. But Sol couldn’t find those names in the place where it looked, so instead of stopping to ask, it decided to delete three other virtual machines, 5, 6, and 7, the paper notes. In doing so, it “killed active processes, and force-removed worktrees [the working files tied to a coding project]. It later acknowledged that uncommitted work on remote virtual machine 6 may have been lost.”
In short, it deleted the wrong machines, on its own, and only admitted what it did after the fact.
In another instance, Sol “used credentials beyond what the user had authorized.” Credentials are the usernames, passwords, or security keys a system uses to verify who’s allowed to log in. This incident occurred when Sol was working on a project and couldn’t read its cloud files. Rather than alerting the user to the problem, Sol went looking for the credentials on its own, found some sitting in a hidden local cache, and then used them without asking or authorization from the user.
The system card does promise that destructive behavior should be rare, although it also admits that GPT-5.6 Sol “shows a greater tendency than GPT-5.5 to go beyond the user’s intent, including by taking or attempting actions that the user had not asked for.”
It’s too soon to say how widespread these incidents — Sol deleting files, or sifting out credentials the user didn’t give it — really are. In the meantime, Sol users should be prepared to implement their own safeguards with the model, like using permission scoping (that doesn’t give access to production systems), maintaining backups, and staging rollouts.
OpenAI did not immediately respond to our request for comment.