DRD 60:I Let AI Claude Access Into My Computer. Here's What Actually Happened.

When people told me AI could access my computer and organise my files, my first thought wasn't "wow, how efficient."

It was "what if it deletes something?" And right behind that — "what if it leaks my data?"

These aren't irrational fears. You've worked hard to build your digital life. Your patient records. Your research notes. Your personal photographs. The idea of handing an AI the keys to all of that feels like leaving your front door open and walking away.

So I didn't give it full access.

I gave it one folder.

The Experiment: One Folder, 400 Unnamed Files

My laptop's screenshot folder had been quietly becoming a disaster for months.

You know how it is. You take a screenshot in a hurry — a form, a slide, a result, a reference. You move on. Then another. Then another. Before you know it, you have 439 files all named something like Misc_Screenshot_2026_04_08_114832.png and absolutely no idea what's inside any of them.

I had screenshots from my Transfusion Medicine work. MBA slides from BITS Pilani. TRAM programme materials. LinkedIn content ideas. Personal stuff. All in one pile, unnamed, unsorted.

I asked Claude Cowork to sort it.

What happened next genuinely surprised me.

Click here to read more about AI & Privacy.

What Is Claude Cowork?

Before I get to the results — a quick explanation for anyone who hasn't heard of this.

Claude is an AI assistant made by Anthropic. Think of it as a very capable thinking partner you can have a conversation with. It can write, analyse, explain, plan, and now — with the right tools — it can actually do things on your computer.

Claude Cowork is a desktop application that gives Claude the ability to take actions on your machine. Not just give you advice — actually run commands, read files, rename things, organise folders. It's like giving your assistant not just a voice, but hands.

Anthropic currently offers several Claude models. The most powerful is Claude Opus 4.6. The one powering most everyday tasks — including Cowork — is Claude Sonnet 4.6. It's fast, capable, and handles complex multi-step tasks well.

The key thing to understand is this: you control what it can access. It doesn't automatically get into everything on your laptop. You define the boundaries.

That's exactly what I did.

What Claude Cowork Actually Did

I gave it access to only my Screenshots folder. Nothing else.

Here's what it did, step by step — and you can see this in the screenshots I'm sharing:

Step 1: It Hit a Wall First

The first thing Claude encountered was a problem I didn't even know I had. My screenshots were stored in OneDrive with "Files on Demand" enabled. This meant the files existed as cloud placeholders — they weren't actually downloaded to the laptop yet. Claude couldn't read them to rename them.

It told me exactly what the problem was and exactly how to fix it.

That alone told me something important: this tool doesn't just barrel forward blindly. It thinks, it explains, it waits for you.

Step 2: It Read Through All 439 Files

Once the files were accessible, Claude systematically read through them — 8 at a time, then 12 at a time, then batches of 24. It was building a mental map of what was inside the folder before making any decisions.

No human assistant would do this without complaining. It just did it.

Step 3: It Sorted Everything Into Categories

After scanning the full folder, Claude organised 433 files into 11 meaningful categories:

• Transfusion_Medicine — 69 files

• TRAM — 60 files

• Personal — 160 files

• Teaching_Academic — 37 files

• MBA_BITS_Pilani — 34 files

• Tools_Software — 21 files

• LinkedIn_Content — 12 files

• AI_in_Medicine — 11 files

• Admin_Official — 11 files

• Research_Publications — 13 files

• Blood_Donation_NGO — 5 files

It didn't just dump everything into broad buckets. It recognised the difference between a Transfusion Medicine form and a TRAM leadership slide. It knew what LinkedIn content looked like versus an official admin screenshot.

That level of context awareness, applied to 433 unnamed files, in one session — that's not a small thing.

How This Changes Your Workflow

Think about what just happened here in terms of time and cognitive load.

Manually going through 439 unnamed screenshots — opening each one, deciding what it is, renaming it, putting it somewhere — would have taken the better part of a day. And you'd have to want to do it, which most of us never do. So the folder just keeps growing.

Claude Cowork did it while I had tea.

More importantly, it did it the way I would have wanted it done — not in some generic way, but in a way that mapped directly to the specific domains of my actual life and work.

For healthcare professionals especially, this matters. We carry enormous cognitive loads. Patient care, research, teaching, administration — it all competes for the same brain. Anything that offloads the low-value organisational tasks without requiring you to surrender privacy or control is worth knowing about.

The Honest Pros and Cons

What Works Well

You stay in control of what it can access. Claude doesn't sneak around — it only works within the boundaries you set. It explains what it's doing and why. When it hits an obstacle, it tells you, rather than guessing or silently failing. The results, in this case, were genuinely accurate and contextually intelligent.

What You Should Know

This is a beta product — it's still maturing. Occasionally it will misclassify something; in this run, 218 files initially ended up in a "Misc" pile before Claude caught it and re-sorted them. It required me to download the OneDrive files first — a step I wouldn't have anticipated on my own. And it does require you to install a desktop application and give it defined folder access, which takes a small amount of setup.

The Fear Question

Was my data leaked? No. Were files deleted? No. Did anything go wrong? Nothing serious — and when something was uncertain, it asked.

The fear I had at the start was legitimate. The outcome was not what I feared.

What the Possibilities Look Like

This was one folder with screenshots. That's a small, contained demonstration.

But think about what becomes possible when you apply this kind of tool thoughtfully:

A research folder with hundreds of PDFs, unsorted. Patient education materials scattered across three locations. Conference submission documents mixed in with teaching slides. Old emails you meant to act on. Draft documents with no clear naming convention.

All of this is solvable — with the right boundaries, and the right level of trust built incrementally.

I started with one folder. That's the right way to start.

What I'd Tell Any Healthcare Professional

You don't have to give AI access to everything. You don't have to understand how it works under the hood. You don't have to stop being cautious.

Start small. Pick one folder that's bothering you. One task that's been sitting undone for weeks. Give it exactly that much access and nothing more.

See what happens.

The AI isn't the risk I thought it was. The real risk is continuing to spend your irreplaceable attention on tasks that a machine can handle — while the work that actually needs you goes waiting.