I, Cyborg: A Shallow Dive Into the Security of Brain-Computer Interfaces (BCIs)
Introduction
I decided to research uncommon security topics, and came across some information on advancements in direct human-device interfaces. That’s right, cyborgs.
Now, it could be said that cyborgs have existed in a lo-tech form since the first artificial implant. I would go as far as to say their roots are in the first limb that was replaced with a contrived object. However, the level of technology with which we associate the quintessential concept of cyborgs as electronically enhanced humans is actually becoming reality. Companies are developing ways to link the human neurologic system directly with external devices.
What’s more, this actually goes beyond the concept of cybernetic enhancements. This is straight-up mind control (no, not that kind). These breakthroughs allow for the manipulation of devices through thought, by connecting with, and reading directly from, the brain and its signals.
Now, before I continue, it is important to note that this technology currently does not read thoughts. At the moment the technology relies on “machine learning algorithms that have been trained to recognize brain activity in the form of electrical impulses and make inferences about emotional states, actions, and expressions.”¹
Okay, so it’s a sort of jury-rigged thought control, but it is still very cool, and as is the case with all technology, there are risks. Not only is this technology being used to manipulate devices, but it is also being used to manipulate and stimulate the brain.
However, this technology will eventually advance, and as Elon Musk indicated when discussing his startup Neuralink, the future includes not only the ability to “read” from the brain, but to “write” to it as well.¹
Putting aside any existential dystopian threats such as an authoritarian society where children are implanted with chips that indoctrinate them with fascist propaganda, and thoughtcrime becomes enforceable reality, I will instead focus on the potential technical risks of such advancements.
Brain-Computer Interfaces (BCIs)
The following is a synopsis of an explanation provided by the Future of Privacy Forum:
The term Brain-Computer Interface (BCI) refers to the technology that links brains directly with devices. There are three levels of BCI:
Non-Invasive
Partially Invasive
Invasive
Non-Invasive methods use external sensors such as electroencephalography (EEG) caps. These methods are the least risky but also the least effective.
Partially Invasive, or semi-invasive, methods apply sensors directly to the surface of the brain. These methods are riskier but are more reliable than non-invasive.
Invasive methods use direct implants into the brain. These methods are very risky but are the most effective.¹
When referring to risk above, I am writing specifically about the physical risk to the subject. Obviously, the risk of injury is extremely high with any invasive surgery, and especially so with the brain.
After all, as we all know, the only field more difficult than brain surgery is rocket science. I jest, of course.
Beyond the physical risks, I was curious to learn more about the cyber risks to such technology. I found that even though the concept and its application are cutting edge, the actual technologies in use are common, such as cloud storage, and wireless transmissions.
Risk Assessment
This topic is going to veer slightly into the realm of science-fiction, with just a dash of conspiracy theory. Before you sneer, consider the technology being discussed.
I’ll approach this systematically as I would any new system being introduced into the environment, albeit in a somewhat shorter form than a full blown risk assessment. I’ll provide:
a description of the vulnerable components,
the security requirements for protecting the components,
some potential threats to the components,
potential threat scenarios in which a vulnerability is realized by a threat causing a negative impact, and finally potential controls to reduce or eliminate the risks posed by the threat scenarios.
What are the vulnerable components?
The subject. The human can be harmed physically, emotionally and mentally.
The data. The captured signals are critical to the system and personal to the subject.
The device. The device is made up of physical and electrical parts, requires a power source, and contains soft/firmware. It has physical and logical systems that collect, process, store and transmit data.
Dependencies. The device and its data may require additional infrastructure to function.
What are the security requirements?
The subject must be protected from harm physically, emotionally and mentally.
The data’s confidentiality, integrity and availability must be protected while it is transmitted, stored and processed.
The device’s physical components and its soft/firmware must be protected against intentional and unintentional corruption, failure and misuse.
The system dependencies must be protected from intentional and unintentional corruption, failure and misuse.
What are some potential threats?
Threat 1: A malicious actor gains remote access to the soft/firmware, or the infrastructure.
Threat 2: The device loses power.
Threat 3: The device’s material is toxic.
Threat 4: The device overheats or releases an electrical charge.
Threat 5: Malicious code is introduced into the device.
Threat 6: The communication between a device and its infrastructure is sniffed.
What are some potential threat scenarios?
Scenario 1: A malicious actor gains access to the data storage and steals data.
Scenario 2: The device loses power, causing its functionality to cease and harm to the subject.
Scenario 3: A device contains toxic chemicals that cause harm to the subject.
Scenario 4: A flaw in the device’s code or structure causes a malfunction that physically harms the subject.
Scenario 5: Malware is introduced into the device that causes data corruption.
Scenario 6: A malicious actor captures traffic between the device and its infrastructure, leading to data disclosure.
What are some potential controls?
Scenario 1: Implement AAA processes to ensure access is restricted to authorized users for authorized uses, and enable activity logging. Encrypt data at rest with the latest encryption standards.
Scenario 2: Implement redundant power sources.
Scenario 3: Implement strict standards for the type of material that may be used.
Scenario 4: Implement AppSec best-practices, such as static code analysis, testing in an isolated environment, and vulnerability / penetration assessments. Stress-test the physical capabilities of devices. Utilize quality components.
Scenario 5: Implement anti-malware capabilities. Protect or eliminate entry points into the device and infrastructure. Segregate data components. Implement immutability.
Scenario 6: Implement and enforce strong encrypted protocols for transmissions.
Though not a complete, nor technologically specific assessment, the above describes some potential risks to the use of BCIs. As BCIs are quite often integrated with other technologies such as cloud storage, neural nets, and machine learning, the risks that are inherent to those systems are also applicable to BCI implementation.
A Closer Look
Given the very specific nature of this technology, I was curious to know if there were more specific controls being utilized. For example, when thinking about the physical connections and information exchanges between the brain and the device, how can that layer be secured? Is it possible to do so? Is it even necessary?
I was not able to discover any technology that satisfied my sci-fi skewed curiosity, but instead began to learn about technologies that are still on the cutting edge, such as fully homomorphic encryption (FHE), which allows for processing encrypted data without decrypting it³, and neural cryptography(neural as in neural net, not actually encrypting the brain), which is the use of neural networks to generate unique keys from neural signals.⁴ Though, not specifically designed for BCI, they are methods for securing BCI data and BCI system dependencies. Anyway, apart from non-invasive methods, the likelihood of a MitM attack between a BCI physical interface and the brain are at this point in time probably impossible without additional invasive surgery, unless the device is already compromised.
Perhaps a more realistic concern is how quickly the advancements in BCI sensitivity and accuracy will allow for remote non-invasive transmission. However, that is veering dangerously close to speculative fiction, and I am attempting to stay within the realm of present reality.
Another concern that steps outside of the technical lines, is regarding ethics, such as the ethics of augmenting humanity, or the potential for exploitation, manipulation and abuse. This is a potentially expansive and sensitive topic, so I will avoid discussing it further at this point. Perhaps I will revisit it in a later article.
Recommendations
Before I chase the rabbit down the what-if-hole, let me provide some realistic suggestions for the current state of BCI technology. The FPF offers five recommendations to help BCI developers address ethical and privacy concerns:
1) Employ Privacy Enhancing Technologies - Number one suggests the use of privacy enhancing technologies (PETs) such as differential privacy, data minimalization and privacy by design.
2) Ensure On/Off User Controls - Number two suggests that users be allowed to switch the device on or off, when an always-on state is not required for functionality.
3) Enshrine Purpose Limitation - Number three suggests that clear statements should be given as to the specific use of the data, user consent requirements, and limits to cross-device collection.
4) Focus on Data Quality - Number four suggests that the most accurate processes and tools available be used to ensure accuracy and precision.
5) Promote Security - Number five suggests that data and the systems that store, process and transmit it are secured against hacking, malware, unauthorized access and tampering, and that data is transmitted using strong encryption.
- Five Top of Mind Data Protection Recommendations for Brain-Computer Interfaces¹
The recommendations are well-conceived and align with the controls identified in my risk assessment, but without regulatory oversight, it will unfortunately be up to organizations whether or not they adopt them. If data can be commoditized or used to enhance features, it will be.
Though the FPF recommendations are good practice, they focus on data privacy. Another area for concern, as mentioned previously, is the risk to the subject. Another source of BCI analysis and suggestions comes from the United States government Accountability Office (USGAO), which offers a list of controls to ensure the health and safety of the subject:
1) Surgical Precision - Number one suggests that highly skilled neurosurgeons equipped with advanced imaging techniques be employed during surgery.
2) Biocompatible Materials - Number two suggests that the device use human-compatible materials to reduce the risk of infection and rejection.
3) Sterile Procedures - Number three suggests strict sterilization procedures during surgery.
4) Encapsulation - Number four suggests that implanted devices are coated to shield them from bodily fluids.
5) Monitoring Systems - Number five suggests the implanted device be continuously monitored to track the device’s performance and the subject’s health, and to detect any complications.
6) Fail-Safe Mechanisms - Number six suggests fail-safes be implemented to isolate or deactive the device in case of malfunction.
7) Post-Op Care - Number seven suggests that the subject be given proper follow-up care and support.
- ² Brain-Computer Interfaces: Applications, Challenges, and Policy Options
Conclusion
The potential health benefits of BCI systems are vast, and the idea that cyborgs as I imagined them are a real possibility, are all very exciting thoughts. Unfortunately, an analysis of the health benefits was outside the scope of this article, so I suggest doing some reading on the subject.
As security is always at the forefront of my mind, I wanted to analyze the potential risks. At the moment, most risks can be addressed with basic cyber hygiene, and standard medical practices, with brilliant minds already working on more advanced protections. I look forward to seeing how this amazing technology advances.
Daily Cuppa
Today’s cup of tea is Tulsi Masala Chai provided by Organic India. A very spicy blend that fills the body and spirit with energy.
References
¹Five Top of Mind Data Protection Recommendations for Brain-Computer Interfaces
Jeremy Greenberg, Katelyn Ringrose - Future of Privacy Forum
https://fpf.org/blog/five-top-of-mind-data-protection-recommendations-for-brain-computer-interfaces/
²Brain-Computer Interfaces: Applications, Challenges, and Policy Options
US Government Accountability Office (USGAO)
https://www.gao.gov/products/gao-25-106952
https://www.gao.gov/assets/gao-25-106952.pdf
³https://engineering.nyu.edu/news/encryption-breakthrough-lays-groundwork-privacy-preserving-ai-models
⁴https://www.cs.ndsu.nodak.edu/~siludwig/Publish/papers/NaBIC2017.pdf
If you enjoyed this article or the site in general, you can buy the author a cup of tea to show your support.