Photo: Charles Thatcher/Getty Images
When Scott Erven was given free rein to roam through all of the
medical equipment used at a large chain of Midwest health care
facilities, he knew he would find security problems–but he wasn’t
prepared for just how bad it would be.
In a study spanning two years, Erven and his team found drug infusion
pumps–for delivering morphine drips, chemotherapy and antibiotics–that
can be remotely manipulated to change the dosage doled out to patients;
Bluetooth-enabled defibrillators that can be manipulated to deliver
random shocks to a patient’s heart or prevent a medically needed shock
from occurring; X-rays that can be accessed by outsiders lurking on a
hospital’s network; temperature settings on refrigerators storing blood
and drugs that can be reset, causing spoilage; and digital medical
records that can be altered to cause physicians to misdiagnose,
prescribe the wrong drugs or administer unwarranted care.
Erven’s team also found that, in some cases, they could blue-screen
devices and restart or reboot them to wipe out the configuration
settings, allowing an attacker to take critical equipment down during
emergencies or crash all of the testing equipment in a lab and reset the
configuration to factory settings.
“Many hospitals are unaware of the high risk associated with these
devices,” Erven says. “Even though research has been done to show the
risks, health care organizations haven’t taken notice. They aren’t doing
the testing they need to do and need to focus on assessing their
risks.”
Erven works as head of information security for Essentia Health,
which operates about 100 facilities–including clinics, hospitals and
pharmacies–in Minnesota, North Dakota, Wisconsin and Idaho. Essentia
decided to open its facilities to a full-scale evaluation in 2012, and
in a remarkable and laudable move, allowed Erven to publicly reveal some
of his findings.
“Many hospitals are unaware of the high risk associated with these devices.”
—Scott Erven
Erven won’t identify specific product brands that are vulnerable
because he’s still trying to get some of the problems fixed. But he said
a wide cross-section of devices shared a handful of common security
holes, including lack of authentication to access or manipulate the
equipment; weak passwords or default and hardcoded vendor passwords like
“admin” or “1234″; and embedded web servers and administrative
interfaces that make it easy to identify and manipulate devices once an
attacker finds them on a network.
Although Erven and his team don’t know whether any of these devices
are connected directly to the internet–they plan a subsequent test to
determine this–many of them are connected to internal networks
accessible via the internet. Hackers could gain access to the devices by
infecting an employee’s computer via a phishing attack, then exploring
the internal network to find vulnerable systems. A hacker who happens to
be in the hospital could also simply plug his laptop into the network
to discover and attack vulnerable systems.
“There are very few [devices] that are truly firewalled off from the
rest of the organization,” he says. “Once you get a foothold into the
network … you can scan and find almost all of these devices, and it’s
fairly easy to get on these networks.”
Everything Was Tested, And Most Of It Was Hackable
Erven, who plans to present some of his findings today at Thotcon in
Chicago, began his research after a security consultancy performing a
penetration test on an Essentia Health network discovered some devices
connected to the network that had security issues. This, combined with
previous research done by other security experts showing problems with
insulin pumps,
defibrillators and
hardcoded passwords in medical devices, prompted Essentia to take an extensive look at all of its equipment.
“We had management backing to see what our risk exposure is across
all health care systems,” he says. “We tested every single device in our
environment–various radiology stuff and MRIs, ultrasound and
mammography systems, cardiology, oncology. We tested all of our lab
systems, surgery robots, fetal monitoring, ventilators, anesthesia.”
One of the main problems they found lay with embedded web services
that allow devices to communicate with one another and feed digital data
directly to patient medical records.
“A lot of the web services allow unauthenticated or unencrypted
communication between the devices, so we’re able to alter the info that
gets fed into the medical record … so you would get misdiagnosis or get
prescriptions wrong,” he says. “The physician is taught to rely on the
information in the medical records … [but] we could alter the data that
was feeding from these systems, due to the vulnerabilities we found.”
Erven says an attacker can collect data passing from medical devices
to patient records, then replay it so that the same data gets passed
into other records.
They also found problems with refrigeration systems for blood and pharmaceutical storage and cryogenics that aren’t protected.
“They all have a web interface that allow you to set the temperature
range,” he says. Although he says the systems include email alerts and
wireless pagers that notify lab and hospital staff if the temperature
falls outside certain boundaries, the systems are only protected by
hardcoded passwords, and once in the system, an attacker can turn off
the email pager notification features or alter the settings to change
when an alert is sent.
Storage systems for X-rays and other images were equally vulnerable.
Erven says the images are generally backed up in centralized storage
units that require no authentication to access. While some of the
front-end systems that physicians and other staff use to access the
images do use hardcoded passwords and log who accesses the images, Erven
says the backup is completely unprotected “and there is no logging if
you go in the backdoor way and grab those images.”
They also found surgery robots connected to internal networks.
Although the robots generally have software firewalls to block
connections to them, Erven and his team found that simply running an
off-the-shelf vulnerability scanner against the firewall caused it to
turn off and fail open.
“But we haven’t figured out yet what we can do once those fail open,” he says.
The Worst Problems
Some of the most disturbing problems they found involved infusion
pumps, ICDs (implantable cardiovascular defibrillators that deliver
shocks to a patient who shows signs of going into cardiac arrest) and CT
scans. They found a number of infusion pumps that have a web
administration interface for nurses to change drug dosage levels from
their workstations. Some of the systems are not password-protected,
while others have hardcoded passwords that are weak and universal to all
customers.
With the CT scan, they could alter configuration files and change
radiation exposure limits that set the amount of radiation patients
receive.
Though targeted attacks would be difficult to pull off in most cases
they examined, since hackers would need to have additional knowledge
about the systems and the patients hooked up to them, Erven says random
attacks causing collateral damage would be fairly easy to pull off.
That’s not the case with implantable defibrillators, however, which could be targeted.
“We found a couple of defibrillator vendors that use a Bluetooth
stack for writing configurations and doing test shocks [against the
patient] when they’re implanted or after surgery,” he says. “They have
default and weak passwords to the Bluetooth stack so you can connect to
the devices. It’s a simple password like an iPhone PIN that you could
guess very quickly.”
A fictional defibrillator attack had a prominent role in an episode of the TV show
Homeland in 2012 but the risks of such an attack are real. Physicians for former Vice President Dick Cheney
had the wireless capability of his defibrillator disabled in 2007 to prevent terrorists from conducting such an attack to kill him.
Although the picture of hospital equipment that Erven and his team
uncovered was gloomy, there was one bright spot among all the bad news —
anesthesia equipment and ventilators are generally not networked and
don’t allow web administration, so someone would have to have physical
access to the devices to alter them.
Hospitals Are Unaware of the Dangers
Erven says that the health care industry is just now waking up to the
security problems with medical equipment, and that the problems exist
because medical equipment has only ever been regulated for reliability,
effectiveness and safety, not for security.
“The vendors don’t have any types of security programs in place, nor
is it required as part of pre-market submission to the [Federal Drug
Administration],” Erven notes. “There’s no security assessment before it
goes to market.”
Last spring, the FDA and DHS issued a notice to the health care industry about
problems with hard-coded passwords in medical devices after
two researchers found them in about 300 medical devices, including ventilators, pumps, defibrillators and surgical and anesthesia devices.
The alert advised health care facilities to examine their systems for
problems and put controls in place to protect them from unauthorized
users. But Erven says health care facilities can only do so much to
wall-off devices; vendors must do more to secure the devices with
encryption and authentication before they sell them to customers and fix
the ones that are already in the field. FDA guidelines for medical
devices now place the onus on vendors to ensure that their systems are
secure and patched, and customers should demand they do so.
Although vendors often tell customers they can’t remove hard coded
passwords from their devices or take other steps to secure their systems
because it would require them to take the systems back to the FDA for
approval afterward, Erven points out that the FDA guidelines for medical
equipment includes a
cybersecurity clause that allows a post-market device to be patched without requiring recertification by the FDA.