Medical devices
The medical device industry designs, makes and wins approval for the products other than medicines that are used in care, from implants and surgical tools to monitors, imaging machines, lab tests and the software that runs them.
1896
Months after Wilhelm Röntgen reported X-rays, the instrument maker Reiniger, Gebbert & Schall of Erlangen, Germany, a forerunner of Siemens Healthineers, was making X-ray tubes designed for medical use, and Röntgen himself wrote to order two. In 1897 the firm advertised them in what Siemens Healthineers calls the world's first catalog of X-ray equipment, and by 1898 it employed three times as many people as before the discovery.
39
Milestones achieved across 6 domains.
8
To watch for: announced, not yet confirmed outside.
12
Open deliverables. Nobody has shipped them.
In one minute
Where it started
Devices became a factory product in 1896, when instrument makers in Germany turned the newly found X-rays into tubes and machines sold to doctors. Electronics then made implants possible: in 1957 Earl Bakken built a battery pacemaker from transistors in four weeks. Rules came after harm. The US began reviewing devices before sale in 1976, after the Dalkon Shield contraceptive injured women, and the EU set one rulebook for devices in 1993.
Where it stands
MedTech Europe, the European trade body, estimates the European market at about €173 billion in 2025, a quarter of the world's, behind the US at 46%. Rules are moving closer together: since 2 February 2026 US factory rules follow ISO 13485, the international quality standard, and in May 2026 the EU made the first four parts of its device database mandatory. Software is now a large part of the trade, and the FDA's list of AI-enabled devices it has authorized had 1,614 entries in September 2026.
What is open
Many open jobs here are checks on public records: which tests stand behind each US clearance, how early adverse-event reports could have warned of each serious recall, and how device plastics change under each sterilization method. AI models that read long PDFs and quote their sources make these jobs possible for a small team. Others are bench work that never touches a person, such as a low-cost robot arm that learns to join tubing and a machine-checked proof that open-source insulin-dosing code keeps to its limits.
Domains
AchievedWatch forOpen
Domain
Progress
Last achieved
Watch for
Open
Design and prototyping
Turning a clinical need into a device design, building and testing early versions, and keeping a record of each decision that someone else can check.
5 · 1 · 2
Last achievedA trial run in software2018
Watch forAgents that check design files
2
openMaterials and sterilization
Choosing what a device is made of so the body tolerates it, and killing the microbes on it, before first use and between uses, without damaging it.
7 · 1 · 2
Last achievedBiocompatibility judged by risk2026
Watch forEPA moves to repeal limits
2
openManufacturing
Manufacturing is making devices in volume, the same way every time, in plants that regulators inspect.
6 · 1 · 2
Last achievedHospital-made implant cleared2026
Watch forOptical inspection of implants
2
openSoftware in devices
This is the code that runs inside devices, or that is a device by itself, and the rules for writing, changing and protecting it.
7 · 2 · 2
Last achievedUpdates cleared in advance2024
Watch forA chat model in dosing
2
openTesting and approval
Before a device can be sold, its maker must show a regulator, or a firm the regulator appoints, that it is safe and does what it claims, using bench tests, lab and animal studies and sometimes studies in people.
7 · 2 · 2
Last achievedChina's fast lane for devices2025
Watch forLighter EU device rules
2
openTracking after sale
After a device is sold, makers, hospitals and regulators collect reports of faults and harm, trace devices by code, and recall or correct them when a problem is found.
7 · 1 · 2
Last achievedA third of reports late2025
Watch forEU incident data in EUDAMED
2
open
Narrower industries
Figureheads
1 person
Advances to build with
4 research notes

NVIDIA GR00T 2
Five demonstrations per task
By NVIDIA's own test, its robot model GR00T 2 averages 95% success after five demonstrations of a task, on 50 tasks picked from those it learns that way.

Gemini 4 Carbon
Cyber defenders first
Google says cyber defenders got Gemini 4 Carbon first, and its 88.2% flaw-finding score is Google's own run on a 183-task subset with guardrails off.

Muse Ember
The jobs agents repeat
Meta's Muse Ember matched Claude Haiku 5.5 on an outside firm's tests at a third of the cost per task, though both list the same prices for text in and out.

Cursor Agent SDK
Findings that cite notes
Cursor released a kit for building AI agents, and its example, a code reviewer that cites the note behind each finding, is known only from Cursor's account.
Open in medical devices
3 of 12 open deliverables
Open · Design and prototyping
Planted errors in design files
A complete, public design file for a simple device that never touches a person, such as a bench pump that moves only water, with user needs, requirements, a risk analysis, drawings, verification tests and their results, all linked to one another. Then a second copy with 50 planted errors of known kinds (a requirement with no test, a test that checks the wrong limit, a hazard with no control, a drawing at the wrong revision), kept as an answer key for scoring AI reviewers and people.
Built with what exists
the open pain-pump requirements as a starting point; OpenRegulatory's templates for ISO 14971 risk files and ISO 13485 records; a syringe-pump prototype built from off-the-shelf parts and run only with water; Git to version every document; Claude and Gemini as reviewers, set to cite the requirement or clause behind each finding, as the code reviewer in the Note "Findings that cite notes" does, and run five times each, because one run can miss what another catches.
Open · Materials and sterilization
An open sterilization damage table
A public, machine-readable table of how device plastics, rubbers and adhesives change after each sterilization method (steam, EtO, gamma rays, electron beam, vaporized hydrogen peroxide): material grade, method, dose or cycle, property measured, change in percent, and the paper, table and page it came from.
Built with what exists
the Europe PMC and OpenAlex APIs to find papers; Claude or Gemini to pull tables from PDFs into a fixed format with page citations, an advance that turns weeks of hand copying into checking; two people to check samples; DuckDB and a public repository such as GitHub or Zenodo; for spot checks, standard plastic test bars sent through a contract sterilizer and pulled on a university tensile tester.
Open · Manufacturing
Teach a robot tubing joints
A low-cost robot arm that joins plastic medical tubing to a connector: it picks up a cut length of tube, pushes it onto the connector, applies a UV-curing adhesive and cures it under a lamp. It learns the job from recorded demonstrations. The joints are test pieces only and are never used on anyone.
Built with what exists
Hugging Face's LeRobot library and its low-cost SO-101 arm; open robot policies such as ACT or SmolVLA, or NVIDIA's GR00T models, which learn a task from a few dozen demonstrations (see the Note "Five demonstrations per task", the Note "No download" and the Note "A real production line"); PVC tubing and plastic luer connectors bought as parts; a UV lamp; a digital force gauge on a manual test stand.
