A close-up of a computer screen displaying complex MRI data being processed by Claude Code. (Illustrative AI-generated image).
- 'Claude, What Do You See?' – The Setup
- How I Fed an MRI to an AI (and Why That's Weird)
- 404 Upvotes and 524 Comments: What the Hacker News Crowd Said
- The Big Picture: Is This Safe? (Short Answer: No.)
- Who Else Is Using AI for Medical Second Opinions?
‘Claude, What Do You See?’ – The Setup
I sat at my laptop, staring at the screen. A few minutes earlier, I had uploaded a file I never thought I’d feed into an AI: my own MRI scan.
The image showed a cross-section of my lower back. A disk herniation, according to my doctor. But I wanted a second opinion. Not from another radiologist – from Claude Code, Anthropic’s popular AI coding tool.
I typed my question: “What do you see here? Any abnormalities?” Then I hit enter.
The response came back in seconds. It described the anatomy, pointed out the herniated disk, and even suggested possible symptoms. It sounded confident. But I wasn’t.
I’m not a doctor. I’m a journalist who got curious after reading a blog post by someone named Antoine. He had done the same thing, and his write-up went viral on Hacker News: 404 upvotes, 524 comments. I wanted to see for myself.
How I Fed an MRI to an AI (and Why That’s Weird)
Claude Code is designed for coding. It helps programmers write and debug software. But it can also process images, including medical scans. That’s technically possible, but it’s not what it was made for.
I converted my MRI file into a standard image format and uploaded it into Claude’s interface. Then I asked questions like a patient grilling a doctor. “Is this disk normal?” “Could that dark spot be something else?”
The AI answered each one. It used medical terms I had to look up. It never hesitated. It gave me a list of possible conditions and recommended follow-up tests. It even added a disclaimer: “I’m an AI and not a doctor. Consult a physician.”
But still, the experience felt weird. Here I was, trusting a piece of software with something as personal as my spine. I felt a mix of awe and unease.
What makes this even weirder is that Claude Code, like most AI models, is not trained specifically on medical images. It learns from text and code. Yet it still gave plausible answers. That’s because it can recognize patterns and has absorbed tons of medical information from the internet. But it doesn’t truly understand – it’s just predicting the next word.
404 Upvotes and 524 Comments: What the Hacker News Crowd Said
When Antoine posted his experience on Hacker News, the community exploded. Four hundred and four upvotes. More than five hundred comments. The discussion reveals a lot about how people feel about AI in medicine.
Many commenters were excited. They saw this as a glimpse into the future. “This is exactly the kind of thing AI should be used for,” one wrote. Another said, “Imagine having an AI assistant that never gets tired and can review thousands of scans.”
But plenty of others warned against it. “You’re playing with fire,” said another comment. “AI can make mistakes that look convincing. This could lead people to delay real treatment.”
The thread showed a split between hope and caution. Some shared their own stories of using AI for health questions. Others pointed out that even if the AI is right most of the time, the one time it’s wrong could be disastrous.
Several commenters asked about the accuracy of the AI’s analysis. No one had verified it. The author hadn’t provided a comparison with a radiologist’s report. That made people uneasy.
What came through clearly was a lack of trust – not in the technology per se, but in how it was being used. People wanted guardrails.
The Big Picture: Is This Safe? (Short Answer: No.)
Let me be clear: using an AI like Claude Code for medical diagnosis is not safe. At least not right now. No AI tool has been approved by the FDA for reading MRIs. Not Claude, not GPT-4, not any of them.
Radiologists spend years training to interpret scans. They learn to spot subtle patterns that an AI might miss. They also understand context – your age, your symptoms, your medical history. An AI just sees pixels.
Even if the AI gets the diagnosis right, it might not communicate the severity correctly. Or it might give a false sense of security. Or it might miss something entirely.
In my case, my actual doctor later confirmed the AI’s findings were on the right track. But that’s just luck. For someone else, the AI could be dead wrong.
There’s also the question of privacy. Uploading a medical scan to a third-party AI means your data leaves your control. Even if the company promises not to use it for training, you’re still sending sensitive info over the internet.
Anthropic, the company behind Claude, hasn’t claimed that Claude Code is for medical use. In fact, they likely discourage it. But they also can’t stop people from trying.
Who Else Is Using AI for Medical Second Opinions?
It turns out, I’m not alone. A lot of people are turning to AI for health advice. The New York Times recently published a list of 21 ways people use AI at work. Some of them involve health: summarizing medical records, suggesting treatment options, even analyzing X-rays.
There are documented cases online of people feeding their blood test results into ChatGPT and asking for interpretations. Others have used AI to read their sleep study reports. The trend is growing.
But these are all personal experiments. No one knows how often these AI analyses are correct. There’s no systematic study. It’s the wild west of consumer health AI.
Meanwhile, companies like Anthropic are making AI more accessible. Claude has been called “wildly popular” by The Week. The Wall Street Journal reported that even non-technical people are blown away by it. That means more people might try similar experiments.
And with the rise of AI coding tools like Claude Code, the barrier is low. You don’t need to be a programmer to upload an image and ask a question. Anyone can do it.
What Anthropic Thinks About All This
Anthropic has not officially approved Claude Code for medical use. In fact, their terms of service likely prohibit it. But the company has focused on making AI safer and more transparent.
They recently published research on “Natural Language Autoencoders” – a way to turn Claude’s internal thoughts into readable text. This is meant to help understand why the AI makes certain decisions. That kind of transparency is crucial for medical applications.
But even with better transparency, the AI is still not reliable enough for diagnosis. Anthropic themselves would probably say: don’t use our tool as a doctor.
Still, the fact that people are using it this way shows a gap in the market. Many people want quick, cheap medical advice. They’re frustrated with slow healthcare systems or high costs. And AI offers a tempting alternative.
What Happens Next: Regulation, Reality, Hope
So where does this leave us? The reality is that AI will likely play a big role in healthcare in the future. But we need better regulation first.
Right now, there are few guardrails for consumer AI in health. The FDA has cleared some AI tools for specific tasks, like detecting breast cancer on mammograms. But those are purpose-built and tested. They’re not general-purpose chatbots.
If people keep using tools like Claude Code for medical advice, someone will eventually get hurt. That could lead to lawsuits, stricter rules, or a backlash against AI altogether.
But there’s also hope. Imagine a future where you can get an accurate, safe second opinion from an AI – after it has been rigorously tested and approved. That could be a game-changer for people who can’t afford or access specialists.
Until then, experiments like mine and Antoine’s are useful as cautionary tales. They show both the promise and the limits of the technology.
After my little test, I didn’t change my treatment plan. I still listened to my doctor. But the AI did make me curious: could it have seen something my doctor missed? Maybe. But I’m not willing to bet my health on it.
And that’s the real question: when it comes to your body, how much risk are you willing to take with an AI that might be wrong?
Frequently Asked Questions
What did the author use an AI tool to analyze?
The author used an AI coding tool called Claude Code to analyze their own MRI scan of their lower back. They wanted to see if the AI could identify abnormalities, specifically a herniated disk.
What is Claude Code designed for?
Claude Code is primarily designed to help programmers write and debug software. However, it also has the capability to process images, which is how the author was able to use it for their MRI scan.
How did the author feed the MRI to the AI?
The author converted their MRI file into a standard image format. They then uploaded this image into Claude's interface and asked questions about the scan, similar to how one might ask a doctor.
Why is using AI for medical scans like MRIs considered weird or not safe?
It's considered weird because AI tools like Claude Code are not specifically trained on medical images and lack the nuanced understanding of a trained radiologist. It is not considered safe because no AI tool is FDA-approved for reading MRIs, and they can make mistakes or miss crucial details.
What was the reaction on Hacker News to a similar experiment?
The reaction on Hacker News was mixed, with many commenters excited about the future potential of AI in medicine. However, others expressed significant caution, warning about the dangers of AI making convincing mistakes and potentially leading people to delay proper medical treatment.
What are the privacy concerns with uploading medical scans to AI?
Uploading a medical scan to a third-party AI means your sensitive personal data leaves your control. Even if the company promises not to use it for training, the information is still being sent over the internet, raising privacy issues.
Does the company behind Claude Code endorse its use for medical diagnosis?
No, Anthropic, the company that created Claude, has not officially approved Claude Code for medical use. Their terms of service likely prohibit such use, and they focus on making AI safer and more transparent, not for medical diagnosis.
