Manufacturing medical devices with cybersecurity firmly in mind is an endeavor that, according to Christopher Gates, an increasing number of manufacturers is trying to get right.
Healthcare delivery organizations have started demanding better security from medical device manufacturers (MDMs), he says, and many have implemented secure procurement processes and contract language for MDMs that address the cybersecurity of the device itself, secure installation, cybersecurity support for the life of the product in the field, liability for breaches caused by a device not following current best practice, ongoing support for events in the field, and so on.
“For someone like myself who has been focused on cybersecurity at MDMs for over 12 years, this is excellent progress as it will force MDMs to take security seriously or be pushed out of the market by competitors who do take it seriously. Positive pressure from MDMs is driving cybersecurity forward more than any other activity,” he told Help Net Security.
Gates is a principal security architect at Velentium and one of the authors of the recently released Medical Device Cybersecurity for Engineers and Manufacturers, a comprehensive guide to medical device secure lifecycle management, aimed at engineers, managers, and regulatory specialists.
In this interview, he shares his knowledge regarding the cybersecurity mistakes most often made by manufacturers, on who is targeting medical devices (and why), his view on medical device cybersecurity standards and initiatives, and more.
[Answers have been edited for clarity.]
Are attackers targeting medical devices with a purpose other than to use them as a way into a healthcare organization’s network?
The easy answer to this is “yes,” since many MDMs in the medical device industry perform “competitive analysis” on their competitors’ products. It is much easier and cheaper for them to have a security researcher spend a few hours extracting an algorithm from a device for analysis than to spend months or even years of R&D work to pioneer a new algorithm from scratch.
Also, there is a large, hundreds-of-millions-of-dollars industry of companies who “re-enable” consumed medical disposables. This usually requires some fairly sophisticated reverse-engineering to return the device to its factory default condition.
Lastly, the medical device industry, when grouped together with the healthcare delivery organizations, constitutes part of critical national infrastructure. Other industries in that class (such as nuclear power plants) have experienced very directed and sophisticated attacks targeting safety backups in their facilities. These attacks seem to be initial testing of a cyber weapon that may be used later.
While these are clearly nation-state level attacks, you have to wonder if these same actors have been exploring medical devices as a way to inhibit our medical response in an emergency. I’m speculating: we have no evidence that this has happened. But then again, if it has happened there likely wouldn’t be any evidence, as we haven’t been designing medical devices and infrastructure with the ability to detect potential cybersecurity events until very recently.
What are the most often exploited vulnerabilities in medical devices?
It won’t come as a surprise to anyone in security when I say “the easiest vulnerabilities to exploit.” An attacker is going to start with the obvious ones, and then increasingly get more sophisticated. Mistakes made by developers include:
Unsecured firmware updating
I personally always start with software updates in the field, as they are so frequently implemented incorrectly. An attacker’s goal here is to gain access to the firmware with the intent of reverse-engineering it back into easily-readable source code that will yield more widely exploitable vulnerabilities (e.g., one impacting every device in the world). All firmware update methods have at least three very common potential design vulnerabilities. They are:
- Exposure of the binary executable (i.e., it isn’t encrypted)
- Corrupting the binary executable with added code (i.e., there isn’t an integrity check)
- A rollback attack which downgrades the version of firmware to a version with known exploitable vulnerabilities (there isn’t metadata conveying the version information).
Overlooking physical attacks
Physical attack can be mounted:
- Through an unsecured JTAG/SWD debugging port
- Via side-channel (power monitoring, timing, etc.) exploits to expose the values of cryptographic keys
- By sniffing internal busses, such as SPI and I2C
- Exploiting flash memory external to the microcontroller (a $20 cable can get it to dump all of its contents)
Manufacturing support left enabled
Almost every medical device needs certain functions to be available during manufacturing. These are usually for testing and calibration, and none of them should be functional once the device is fully deployed. Manufacturing commands are frequently documented in PDF files used for maintenance, and often only have minor changes across product/model lines inside the same manufacturer, so a little experimentation goes a long way in letting an attacker get access to all kinds of unintended functionality.
No communication authentication
Just because a communications medium connects two devices doesn’t mean that the device being connected to is the device that the manufacturer or end-user expects it to be. No communications medium is inherently secure; it’s what you do at the application level that makes it secure.
Bluetooth Low Energy (BLE) is an excellent example of this. Immediately following a pairing (or re-pairing), a device should always, always perform a challenge-response process (which utilizes cryptographic primitives) to confirm it has paired with the correct device.
I remember attending an on-stage presentation of a new class II medical device with a BLE interface. From the audience, I immediately started to explore the device with my smartphone. This device had no authentication (or authorization), so I was able to perform all operations exposed on the BLE connection. I was engrossed in this interface when I suddenly realized there was some commotion on stage as they couldn’t get their demonstration to work: I had accidentally taken over the only connection the device supported. (I then quickly terminated the connection to let them continue with the presentation.)
What things must medical device manufacturers keep in mind if they want to produce secure products?
There are many aspects to incorporating security into your development culture. These can be broadly lumped into activities that promote security in your products, versus activities that convey a false sense of security and are actually a waste of time.
Probably the most important thing that a majority of MDMs need to understand and accept is that their developers have probably never been trained in cybersecurity. Most developers have limited knowledge of how to incorporate cybersecurity into the development lifecycle, where to invest time and effort into securing a device, what artifacts are needed for premarket submission, and how to proper utilize cryptography. Without knowing the details, many managers assume that security is being adequately included somewhere in their company’s development lifecycle; most are wrong.
To produce secure products, MDMs must follow a secure “total product life cycle,” which starts on the first day of development and ends years after the product’s end of life or end of support.
They need to:
- Know the three areas where vulnerabilities are frequently introduced during development (design, implementation, and through third-party software components), and how to identify, prevent, or mitigate them
- Know how to securely transfer a device to production and securely manage it once in production
- Recognize an MDM’s place in the device’s supply chain: not at the end, but in the middle. An MDM’s cybersecurity responsibilities extend up and down the chain. They have to contractually enforce cybersecurity controls on their suppliers, and they have to provide postmarket support for their devices in the field, up through and after end-of-life
- Create and maintain Software Bills of Materials (SBOMs) for all products, including legacy products. Doing this work now will help them stay ahead of regulation and save them money in the long run.
They must avoid mistakes like:
- Not thinking that a medical device needs to be secured
- Assuming their development team ‘can’ and ‘is’ securing their product
- Not designing-in the ability to update the device in the field
- Assuming that all vulnerabilities can be mitigated by a field update
- Only considering the security of one aspect of your design (e.g., its wireless communication protocol). Security is a chain: for the device to be secure, all the links of the chain need to be secure. Attackers are not going to consider certain parts of the target device ‘out of bounds’ for exploiting.
Ultimately, security is about protecting the business model of an MDM. This includes the device’s safety and efficacy for the patient, which is what the regulations address, but it also includes public opinion, loss of business, counterfeit accessories, theft of intellectual property, and so forth. One mistake I see companies frequently make is doing the minimum on security to gain regulatory approval, but neglecting to protect their other business interests along the way – and those can be very expensive to overlook.
What about the developers? Any advice on skills they should acquire or brush up on?
First, I’d like to take some pressure off developers by saying that it’s unreasonable to expect that they have some intrinsic knowledge of how to implement cybersecurity in a product. Until very recently, cybersecurity was not part of traditional engineering or software development curriculum. Most developers need additional training in cybersecurity.
And it’s not only the developers. More than likely, project management has done them a huge disservice by creating a system-level security requirement that says something like, “Prevent ransomware attacks.” What is the development team supposed to do with that requirement? How is it actionable?
At the same time, involving the company’s network or IT cybersecurity team is not going to be an automatic fix either. IT Cybersecurity diverges from Embedded Cybersecurity in many respects, from detection to implementation of mitigations. No MDM is going to be putting a firewall on a device that is powered by a CR2032 battery anytime soon; yet there are ways to secure such a low-resource device.
In addition to the how-to book we wrote, Velentium will soon offer training available specifically for the embedded device domain, geared toward creating a culture of cybersecurity in development teams. My audacious goal is that within 5 years every medical device developer I talk to will be able to converse intelligently on all aspects of securing a medical device.
What cybersecurity legislation/regulation must companies manufacturing medical devices abide by?
It depends on the markets you intend to sell into. While the US has had the Food and Drug Administration (FDA) refining its medical device cybersecurity position since 2005, others are more recent entrants into this type of regulation, including Japan, China, Germany, Singapore, South Korea, Australia, Canada, France, Saudi Arabia, and the greater EU.
While all of these regulations have the same goal of securing medical devices, how they get there is anything but harmonized among them. Even the level of abstraction varies, with some focused on processes while others on technical activities.
But there are some common concepts represented in all these regulations, such as:
- Risk management
- Software bill of materials (SBOM)
- “Total Product Lifecycle”
But if you plan on marketing in the US, the two most important document should be FDA’s:
- 2018 – Draft Guidance: Content of Premarket Submissions for Management of Cybersecurity in Medical Devices
- 2016 – Final Guidance: Postmarket Management of Cybersecurity in Medical Devices (The 2014 version of the guidance on premarket submissions can be largely ignored, as it no longer represents the FDA’s current expectations for cybersecurity in new medical devices).
What are some good standards for manufacturers to follow if they want to get cybersecurity right?
The Association for the Advancement of Medical Instrumentation’s standards are excellent. I recommend AAMI TIR57: 2016 and AAMI TIR97: 2019.
Also very good is the Healthcare & Public Health Sector Coordinating Council’s (HPH SCC) Joint Security Plan. And, to a lesser extent, the NIST Cyber Security Framework.
The work being done at the US Department of Commerce / NTIA on SBOM definition for vulnerability management and postmarket surveillance is very good as well, and worth following.
What initiatives exist to promote medical device cybersecurity?
Notable initiatives I’m familiar with include, first, the aforementioned NTIA work on SBOMs, now in its second year. There are also several excellent working groups at HSCC, including the Legacy Medical Device group and the Security Contract Language for Healthcare Delivery Organizations group. I’d also point to numerous working groups in the H-ISAC Information Sharing and Analysis Organization (ISAO), including the Securing the Medical Device Lifecycle group.
And I have to include the FDA itself here, which is in the process of revising its 2018 premarket draft guidance; we hope to see the results of that effort in early 2021.
What changes do you expect to see in the medical devices cybersecurity field in the next 3-5 years?
So much is happening at high and low levels. For instance, I hope to see the FDA get more of a direct mandate from Congress to enforce security in medical devices.
Also, many working groups of highly talented people are working on ways to improve the security posture of devices, such as the NTIA SBOM effort to improve the transparency of software “ingredients” in a medical device, allowing end-users to quickly assess their risk level when new vulnerabilities are discovered.
Semiconductor manufacturers continue to give us great mitigation tools in hardware, such as side-channel protections, cryptographic accelerators, virtualized security cores. Trustzone is a great example.
And at the application level, we’ll continue to see more and better packaged tools, such as cryptographic libraries and processes, to help developers avoid cryptography mistakes. Also, we’ll see more and better process tools to automate the application of security controls to a design.
HDOs and other medical device purchasers are better informed than ever before about embedded cybersecurity features and best practices. That trend will continue, and will further accelerate demand for better-secured products.
I hope to see some effort at harmonization between all the federal, state, and foreign regulations that have been recently released with those currently under consideration.
One thing is certain: legacy medical devices that can’t be secured will only go away when we can replace them with new medical devices that are secure by design. Bringing new devices to market takes a long time. There’s lots of great innovation underway, but really, we’re just getting started!