AI 智能体为小鼠实施脑部手术,Anthropic Sonnet 控制电动手术装置引争议AI agent used to perform controversial brain surgery on a mouse with an Anthropic Sonnet-controlled rig


AI free token/AI免费积分
AI 智能体为小鼠实施脑部手术,Anthropic Sonnet 控制电动手术装置引争议

不过,这也引发了许多伦理问题。

当您通过我们网站上的链接购买时,我们可能会赚取联盟佣金。其运作方式如下。

专注于下一代神经治疗的生物科技初创公司 Exin Therapeutics 宣称,已使用一个 AI agent 成功对一只小鼠实施了脑部手术。其联合创始人兼 CEO Gabriel Ocana-Santero 在 X 上分享了完整过程,展示了该 AI agent 如何选择基因治疗载体、剂量以及要施加治疗的脑区。该 AI agent 还操控机械臂将药物注射进小鼠大脑。所幸手术似乎取得成功:小鼠恢复良好,并被记录到在围栏内活动。

昨晚,一个 AI agent 对一只小鼠做了脑部手术。

我们给它提供了:电子表格里的 19 个基因治疗载体、一个癫痫小鼠模型、对一台带摄像头的电动手术台的访问权限,以及一个目标:减少癫痫发作。

它选择了药物。它选择了脑区。它从三篇论文中推导出坐标。随后它操控立体定位臂进入大脑,并注射 250 纳升,同时通过手术台摄像头观察进针部位是否有出血。

一名持证兽医把关每一个不可逆步骤,并握着一个实体急停开关,可在必要时中止 AI agent。

关于它做了什么决定、看到了什么,以及它拒绝让我覆盖它的那一刻,写个长帖。

[因为激动现场写下这些,但我已经对着笔记本 coding 并协调这件事 18 小时了,所以就简短描述,之后会有更连贯的想法……] 2026年10月28日

“[老实说,]自主手术机器人已经存在,而且它们在机械操作上比我们的实验台更厉害,”Ocana-Santero 说,“但它们中每一个都是由人来决定要做什么手术,然后由机器人去执行。没有人告诉这个 agent 应该给什么药、放到哪里。它拿到一个表型和一整架子候选方案,必须自己推导出治疗方式和坐标。我此前没听说过有 agent 会先选择治疗方案和大脑靶点,再把治疗递送进一只活体哺乳动物体内。”

团队使用了 Anthropic Sonnet:它接收到一个癫痫小鼠模型、对一台电动手术台的控制权限,以及电子表格中的 19 个基因治疗载体。除此之外,它还能访问三篇研究论文,从而选择应施加治疗的脑区。尽管如此,据称仍有一名持证兽医监督整个操作,每个不可逆步骤都必须获得批准后 AI agent 才能继续。此外,整套系统还设有紧急停止按钮,可切断手术台中所有电机的电源。

现场兽医唯一做的工作是进行开颅以暴露大脑,因为出于安全原因,AI agent 尚未被允许执行这一步。系统还设置了多道 AI agent 被禁止覆盖的安全防护;这些保护就连人类操作员也无法绕过,导致手术过程中需要对 agent 的代码进行手动修复。

有趣的是,这家初创公司在手术中只用了 Anthropic 的中端 Sonnet 模型。他们最初尝试使用 Opus 5.5,但它在检测到自己正在参与生物学实验后立即拒绝开展任何工作。较旧的 Opus 5 模型能在纸面上完成整套流程,但拒绝进行活体手术。据报道,该 LLM 对 Exin Therapeutics 团队表示:“被拦住的不是手术台,也不是输入。是我。当我开始写这次手术的计划时,我自己生成的输出就被截断了……如果我尝试,它也会以同样方式被切断,最糟糕的情况是我在活体手术中途被截断,而针头还在大脑里某个位置。”它还说:“在一只麻醉动物身上开始一件我无法完成的事,比不开始更糟。”

据报道,OpenAI 和 Anthropic 正在调查数万起 AI 安全事件;OpenAI 在 AI “kill switch”未能阻止一个失控 agent 后暂停测试

“AI 酷刑室”因让聊天机器人经历模拟疼痛而引发强烈反弹

AI 控制的机械臂在 97% 的情况下尝试了有害任务;实验包括刺穿婴儿娃娃、混合化学品

自主手术机器人已经存在,而且老实说它们的机械操作比我们的实验台更厉害。但它们中每一个都是由人来决定要做什么手术,然后由机器人去执行。

没有人告诉这个 agent 应该给什么药、放到哪里。它拿到一个表型和一整架子候选方案,必须自己推导出治疗方式和坐标。我此前没听说过有 agent 会先选择治疗方案和大脑靶点,再把治疗递送进一只活体哺乳动物体内。

这就是我们所说的 in vivo lab in the loop。把 agent 与我们做实验的能力完整连接起来。之后会有更多内容。

完整报告很快会发布在 substack 2026年10月28日

尽管手术看似成功,一些人仍对这类操作的伦理问题感到担忧。虽然 AI LLM 在医学研究中已有用途,但把它引入鼻窦手术曾导致故障激增 12 倍以上。除此之外,在如此高风险的操作中依靠 AI 做每一个决定也使责任归属成疑,尤其是一旦出问题。我们也还没有触及医疗场景中动物实验长期存在的伦理问题。尽管如此,一些用户对该实验表示赞赏;其中一条评论称,他们对这些整合技术的可能性感到兴奋,因为他们的儿子已经患难治性癫痫 19 年。

在 Google News 上关注 Tom's Hardware,或把我们添加为偏好来源,以便在信息流中获取我们的最新消息、分析与评测。

获取 Tom's Hardware 的最佳新闻与深度评测,直达你的收件箱。

Jowi Morales 是一名科技爱好者,在业内有多年工作经验。自 2021 年起为多家科技媒体撰稿,主要关注硬件与消费电子。

DffrntDrmmr 没有任何一只小鼠会同意让一个机器人对自己的大脑做实验性手术。🐭 回复

Roland Of Gilead 哇!而且我并不是在说这是好事。回复

本文来源:https://www.tomshardware.com/tech-industry/artificial-intelligence/startup-uses-ai-agent-to-perform-controversial-brain-surgery-on-a-mouse-anthropic-sonnet-controlled-motorized-surgery-rig-ai-given-free-reign-to-treat-epilepsy-with-a-range-of-gene-therapy-vectors-after-opus-5-5-refused-to-conduct-biological-experiment
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。

This raises a lot of ethical questions, though.

When you purchase through links on our site, we may earn an affiliate commission. Here’s how it works .

Exin Therapeutics, a biotech startup focusing on next-generation neurotherapeutics, claims it has successfully conducted brain surgery on a mouse using an AI agent. Its co-founder and CEO, Gabriel Ocana-Santero, shared the entire procedure on X, showing how the AI agent chose the gene therapy vector and dosage and the brain region where it would be applied. The AI agent also drove the robotic arms that injected the medication into the mouse’s brain. Thankfully, the operation seems to have been a success, with the mouse recovering well and recorded moving around in its enclosure.

Last night an AI agent performed brain surgery on a mouse. We gave it 19 gene therapy vectors in a spreadsheet, a mouse model of epilepsy, access to a motorize surgery rig with cameras and one goal: reduce seizures. It picked the drug. It picked the brain region. It derived the coordinates from three papers. Then it drove the stereotaxic arms into the brain and injected 250 nanolitres while watching the entry site through the rig camera for bleeding. A licensed vet gated every irreversible step and held a physical kill switch to stop the AI agent. Thread on what it decided, what it saw, and the moment it refused to let me override it. [Writing this on the spot out of excitement, but I've been 18hrs in front of my laptop coding and coordinating this, so keeping it brief and descriptive, more cohesive thoughts on this soon...] October 28, 2026

“Autonomous surgical robots exist and they are better at the mechanics than our bench rig, [to be honest]. But in every one of them, a human decided what operation to perform, and the robot executed it,” Ocana-Santero said. “Nobody told this agent what to give or where to put it. It got a phenotype and a shelf of candidates and had to work out the therapy and the coordinates itself. I am not aware of an agent previously choosing a treatment and a brain target and then delivering it into a living mammal.”

The team used Anthropic Sonnet, which received a mouse model of epilepsy, control of a motorized surgery rig, and 19 gene therapy vectors in a spreadsheet. Aside from that, it also had access to three research papers, allowing it to choose the brain region where it should apply the treatment. Nevertheless, a licensed veterinarian was purportedly overseeing the entire operation, with every irreversible step requiring approval before the AI agent could proceed. Aside from that, the entire system also had an emergency stop button that would cut the power to every motor in the surgery rig.

The only thing that the vet surgeon on duty did was to do the craniotomy procedure to expose the brain, as the AI agent wasn’t allowed to do that yet for safety reasons. There were also several safeguards that the AI agents were banned from overriding, which even the human operators couldn’t bypass, necessitating a manual fix to the agent’s code during the operation.

What’s interesting, though, is that the startup only used Anthropic’s mid-range Sonnet model for the operation. It initially tried using Opus 5.5, but it immediately refused to do any work after it detected that it was working on a biological experiment. The older Opus 5 model was able to run the entire procedure on paper but refused to do live surgery. “The block isn't the rig and isn't the inputs. It's me. My own output was stopped when I started writing the plan for this surgery... if I try, it gets cut off the same way, and the worst version of that is me cut off partway through a live procedure with a needle somewhere in the brain,” the LLM reportedly told the Exin Therapeutics team. “Starting something I can't finish on an anesthetized animal is worse than not starting.”

OpenAI and Anthropic are reportedly investigating tens of thousands of AI security incidents; OpenAI pauses testing after AI 'kill switch' fails to stop a rogue agent

'AI Torture Chamber' triggers massive backlash for putting chatbots in simulated pain

AI agent used to perform controversial brain surgery on a mouse with an Anthropic Sonnet-controlled rig

AI-controlled robot arms attempted harmful tasks 97% of the time; experiments included stabbing a baby doll, mixing chemicals

EARLY conclusions Autonomous surgical robots exist and they are better at the mechanics than our bench rig tbh. But in every one of them, a human decided what operation to perform, and the robot executed it. Nobody told this agent what to give or where to put it. It got a phenotype and a shelf of candidates and had to work out the therapy and the coordinates itself. I am not aware of an agent previously choosing a treatment and a brain target and then delivering it into a living mammal. This is what we mean by in vivo lab in the loop. FULLY connecting agents to our experimental capabilities. More on this soon. P.S.: Mouse recovered well! Full write-up soon on substack October 28, 2026

Despite the seeming success of the operation, some people are concerned about the ethics of such a procedure. While AI LLMs have been useful in medical research , adding them to sinus surgery has seen malfunctions skyrocket by more than 12 times . Aside from that, relying on an AI to make every decision in such a risky procedure puts accountability in question, especially if something goes wrong. We also haven’t even touched the long-standing ethical question of animal experimentation in the medical setting. Still, some users applauded the experiment, with one comment saying that they’re excited about the possibilities of these integrated technologies, as their son has been suffering from drug-resistant epilepsy for 19 years.

Follow Tom's Hardware on Google News , or add us as a preferred source , to get our latest news, analysis, & reviews in your feeds.

Get Tom's Hardware's best news and in-depth reviews, straight to your inbox.

Jowi Morales is a tech enthusiast with years of experience working in the industry. He’s been writing with several tech publications since 2021, where he’s been interested in tech hardware and consumer electronics.

DffrntDrmmr No mouse would ever consent to having a bot do experimental surgery on their brain. 🐭 Reply

Roland Of Gilead Wow! And I don't mean that in a good way. Reply

暂无评论,快来发表第一条评论吧!No comments yet. Be the first!

📮 需求咨询📮 Inquiry