Neuralink Performs Its First Brain Implant Without Cutting the Dura

In May 2026, at Toronto's UHN alongside neurosurgeon Andres Lozano, the company drove electrode threads straight through the brain's toughest membrane, using a vein-injected dye and a laser depth-sensor to see and steer where the surgeon is now working blind.

The interesting thing about Neuralink's latest surgery isn't the threads or the robot. It's a subtraction. For every implant so far, the procedure has involved cutting open the dura, the tough leathery sheet that wraps the brain just under the skull, and removing a piece of it so the surgeon can see the cortex and place electrodes. The new version deletes that step outright. The robot now punches electrode threads directly through the intact dura and into the brain. That sounds like a small change. It's actually the move that decides whether this ever reaches millions of people or stays a boutique operation for a few dozen patients a year.

Key Takeaways

• The first transdural Neuralink implant was performed in May 2026 at UHN in Toronto with neurosurgeon Andres Lozano, the most advanced version of the surgery the company has run.
• Skipping the durotomy, the step where surgeons cut and remove part of the dura, is what makes the procedure faster, safer, and less invasive all at once.
• Leaving the dura intact blinds the surgeon to two critical things: where the blood vessels are, and how far down the cortex sits. Neuralink solved both with imaging.
• ICG video angiography injects a dye into a vein, then infrared light makes the blood vessels glow through the leathery membrane so the robot can steer threads around them.
• Optical coherence tomography, a laser-based scanner, measures the distance from the top of the dura to the cortex with enough precision to plant threads at the right depth.
• That depth isn't fixed. In a living person the gap between dura and cortex keeps shifting with the brain's motion, which is exactly why a static measurement wouldn't cut it.
• Neuralink's original needle couldn't reliably pierce the dura, so the team widened its diameter slightly, and built a synthetic dural membrane to run hundreds of puncture tests on the bench.
• The exposed brain in the old procedure was about the size of a quarter. In the new one, the surgeon sees essentially nothing through the dura and relies entirely on the imaging stack.
• The stated endgame is robots running the surgery under clinical supervision, and deleting the durotomy is the setup move that makes automation tractable.

---

The Membrane Neuralink Used to Cut Away

Start with what the dura actually is. It's the outermost of the brain's protective layers, a tough, leather-like sheet sitting right beneath the skull. Nature put it there as armor. Every Neuralink implant to date has treated it as an obstacle to remove: open the skull, open the dura, take out a piece so you can look at the cortex directly, then thread the electrodes in.

Removing that armor has costs. It's another cut, another opening, another surface exposed, more time on the table, more that can go wrong. Preserving the dura and going through it instead collapses all of that into one motion. Safer, faster, less invasive, and easier to repeat. When you're trying to help people with paralysis and other unmet neurological needs, repeatable is the whole point.

Working Blind Through Leather

Here's the problem with keeping the dura closed. The moment you don't cut it, you lose your eyes. With the old opening, the surgeon looked at a patch of exposed brain about the size of a quarter and could see the vasculature to avoid it. Through the intact dura, you see basically nothing. It's opaque leather.

That creates two specific blind spots. You can't see the blood vessels you must not hit, and you can't see how deep the cortex sits below you. Miss on the first and you cause a bleed. Miss on the second and your electrodes land in the wrong place. Solving the transdural surgery was never really about pushing harder. It was about restoring vision the membrane takes away.

Seeing Blood Vessels That Glow

The vessel problem got solved with a technique borrowed from vascular imaging. ICG, indocyanine green, is a dye you inject into a vein. Hit it with infrared light and the vessels light up, glowing right through the dura. Neuralink runs this as video angiography, so the robot gets a live map of where the blood flow is and can route threads safely around it.

I like this because it's not exotic. ICG angiography is established clinical tooling. The engineering was in wiring that visualization into the robot's insertion path in real time, so avoidance isn't a surgeon's judgment call but a constraint the machine plans around.

A Laser That Measures a Moving Target

Depth was the harder half. To know how far the cortex sits below the dural surface, Neuralink built an optical coherence tomography system into the robot head. OCT uses laser light, sent down a fiber to the robot head, bounced off the brain, and captured on the way back, to reconstruct the tissue as a 3D volume. You can resolve the thin dural layer, the subarachnoid space beneath it, and the cortex under that.

The reason this matters more than a single measurement: the target moves. In a living human, the distance from the dura to the cortex is actively changing as the brain shifts and pulses. A number you took a second ago is already stale. OCT lets the robot track that gap continuously and plant each thread at the correct depth anyway. That's the difference between a bench demo and something you'd let near a person.

Why a Slightly Fatter Needle Mattered

Seeing through the dura is only half the job. You still have to get through it. The original needle design simply couldn't reliably penetrate that leathery layer, and "reliably" is the operative word in surgery. The fix was almost boring: widen the needle diameter a little, just enough to carry the electrodes through the membrane cleanly.

I point this out because it's a good reminder of how real hardware progress happens. Not one heroic breakthrough. A dye system, a laser scanner, and a needle that's a hair thicker, all made to work together.

Building a Fake Dura to Break Hundreds of Times

You can't iterate a puncture mechanism on live patients. So before any of this reached an operating room, the team had to build a testing pipeline that didn't exist. The centerpiece was a synthetic dural membrane, a bench-top proxy engineered to match the real thing on the properties that matter: thickness and puncture force, how hard you have to push to get through.

With that stand-in, they ran hundreds of insertion tests. They built a separate brain proxy too, layered to mimic human anatomy and even the brain's motion, to validate the OCT depth-sensing. This is the unglamorous work that decides whether a medical device is trustworthy. Model the tissue accurately, break it a few hundred times in the lab, and only then go near a skull.

Delete the Step Before You Automate It

Now the part that actually generalizes. Neuralink's mission math is simple: help as many people as possible means do as many surgeries as possible, and that eventually means robots performing the procedure under clinical supervision rather than a scarce human surgeon doing each one by hand.

But the smart move came before automation. The question wasn't "how do we automate the durotomy?" It was "can we delete the durotomy entirely?" You don't automate a step you can remove. Every step you delete is one you never have to make a robot do, never have to make safe, never have to make fast. Transdural insertion is that deletion. It's the same first-principles instinct that shows up whenever a company is serious about scale: the best part is no part, the best step is no step.

The Part I'd Watch

I'm genuinely impressed by the engineering here, and I want to be honest about the uncertainty too. This is a first-of-its-kind surgery, done once so far in an ongoing trial. One successful transdural implant tells you the approach is viable. It doesn't yet tell you the failure rate across hundreds of varied human anatomies, where dural thickness and vessel layout differ from person to person.

The thing to track over the next year is consistency. Does the ICG-plus-OCT stack hold up when the anatomy is messy, and does thread placement stay precise once robots run the procedure with a surgeon supervising rather than operating? If it does, Neuralink didn't just make one surgery cleaner. It removed the bottleneck that was standing between a research implant and something you could actually deliver at scale. That's the bet worth watching.