In this paper we explore how users can be uniquely identified using biometrics other than fingerprints, facial recognition, iris recognition etc on a continuous basis. We explore the use to techniques such as typing style, computer use style to see if we can create a model to uniquely identify a user based on the way they type and use the computer. As this method allows a system to constantly reauthenticate a user based on characteristics that are almost impossible to fake we look at the complexity of how this can be integrated as a security measure for secure systems. We also look at the pros and cons of implementing this authentication mechanism and explore potential problems this system generates for the user and administrators. Specifically, we look at how the system would deal with users who are sick, under medication or stress that could change their usage patterns and is it worth the expense and privacy issues to implement such a system.
Introduction and background
User authentication is the process of verifying the identify of a user or process trying to access a system, online service, connected device, infrastructure resources etc. Traditionally authentication is done by having the user provide one or more of the following:
- Something they know
- Something they have
- Something they are
Let’s look at each of these one by one. The oldest way of authentication to computer systems is using usernames and passwords. The first password protection system was implemented in 1961 by Fernando J. Corbató at MIT (Workos, 2020). This allowed the system to identify users based on a secret password that only they knew. The first set of passwords were stored in plain text, but then password encryption was implemented so that users could not read the passwords for other users.
However, passwords can be leaked or guessed. In the past few years there have been major leaks of authentication data which have been decrypted and sophisticated password crackers have been created that can crack passwords based on dictionary attacks and brute force attacks. To safeguard against this attack vector another authentication mechanism was created that authenticates users based on something they have with them. This can include hardware keys, smartcards etc and these hardware devices would contain an embedded certificate that can be used to uniquely identify the holder.
The final method of authentication is something you are, which is provided by Biometric authentication. Some of the biometric methods that can be used are fingerprints, hand geometry, retinal or iris scans, face scans, and voice analysis. Fingerprints, Face Scans and iris scans are the most widely used biometric method in use today.
Multifactor Authentication
When a system uses a combination of one or more of the authentication methods described in the previous section the system is said to be using Multi-factor Authentication (MFA). The key point to remember is that a system is only considered to be using MFA if the authentication factors are in at least two of the categories. So if the authentication mechanism uses a password and a second pin to authenticate, it won’t count as MFA because both are things that you know.
Weaknesses in the current User Authentication methods
The current user authentication methods have several weaknesses that make it easy for attackers to compromise and bypass the checks. Complex passwords are harder to crack or guess than simple passwords, but they are harder for users to remember. So, users tend to use the same passwords across multiple sites or use passwords that are simple to remember. Unfortunately, passwords that are simple to remember are also easy to guess.
Another risk is that an attacker can compromise a site or server using vulnerabilities in the OS, services or applications running on it. Once they have access, they can gain access to the stored passwords for all users and depending on the encryption scheme used the passwords for user accounts can be guessed quickly. This is an attack vector that has been seen frequently over the past few years with password lists for major sites such as LinkedIn (Morris, 2021) and Yahoo (Goel & Perlroth, 2016) etc being compromised and leaked.
Hardware tokens or smart cards can be cloned, copied or stolen. If the card is not deactivated when it is lost or stolen an attacker can use it to gain access to restricted resources. Tools to create copies of smartcards are available easily in the market (Benchoff, 2016) using which an attacker can clone the cards quickly.
Biometrics was touted as an authentication mechanism that is almost impossible to bypass but unfortunately the hype didn’t match reality. Fingerprint authentication systems have been compromised using copies of fingerprints lifted from glasses, door knobs etc transferred to jello, Glycerin and gelatin. (Barral & Tria, 2009)
Facial recognition systems have been fooled by photographs and cosmetics. Researchers have also used the StyleGAN Generative Adversarial Network (GAN) to create master faces that can be used to impersonate 40% of the population. (Shmelkin et al., 2021)
Voice authentication systems have been bypassed using voice recordings and AI based ‘deep fake’ technologies. Amazon recently showcased technology that allows Alexa to impersonate the voices of people based on a few minutes long voice recording of the person being impersonated.
Similar bypasses have been found for all authentication mechanisms in use currently and thus researchers have been exploring new authentication mechanisms which would be harder to bypass and fool. One such field being explored in behavioral biometrics and we will explore the field, it’s implications, the pros and cons of the tech in this paper.
Introduction to Behavioral Biometrics
Behavioral biometrics is the study and use of uniquely identifying and measurable patterns in human activities that can include keystroke dynamics, gait analysis, mouse use characteristics, signature analysis etc. The field postulates that a user can be identified based on these characteristics just as uniquely as they can be using physical biometrics.
Another advantage of using Behavioral Biometrics over physical biometrics is that it doesn’t require specialized equipment to collect the data. Data can be collected using existing hardware and only requires software analysis and processing which makes it cheaper to implement to a certain extent and we will look at this in more detail later in the paper.
Behavioral Biometrics can include the following:
Keystroke Dynamics:
According to the studies, if a group of users is asked to type the paragraph of text, each of them will type the text slightly differently with different delays between each character being typed, and different rhythms for the text. This allows a system to identify the user based on how they type including criteria such as:
- The user’s typing speed
- Time elapsed between each consecutive keystroke
- The time that each key is held down
- The frequency with which the number pad keys are used
- The timing and sequence of the keys used to type a capital letter
- The Error Rate in typing, such as using the Backspace keys and words repeatedly mistyped by the user.
As each person would type the password slightly differently the system can use it to identify the authorized user and block attackers who might have gained the password for a given user.
Cursor Movement:
This uses the tracking speed, clicks and path taken by the mouse cursor movement during use to create a profile for the active user. This would be useful if the user uses the same set of applications frequently, if they are using a varied set of applications that keep changing then this would not be accurate.
Finger pressure on keypad:
This analyses the pressure on the keyboard to create a user profile. This is a lot more relevant for mobile devices and other devices with a touchscreen interface as the allow us to capture pressure details easily without extra hardware.
Posture:
Every person has a different way of standing and a sufficiently trained system can look for differences in how the person sits in front of the computer and their posture while using the system.
Gait:
Gait analysis attempts to identify a person based on their walking style, which includes movements such as stride length, posture, and speed of travel etc.
Each of the methods we listed above can potentially be used to continuously re-validate a logged in user.
Historical use of Behavioral Biometrics for authentication
Historically, behavioral biometrics have been in use since the 1860s when experienced telegraph operators were able to identify individual operators by the way they would send the signals. In World war II allied officers used it to validate the authenticity of messages they received based on how they were sent. (Das, 2020) Similarly, other organizations used this ability as well as an extra validation layer when communicating instructions over telegraph.
The Military has also used gait recognition to identify imposters in their base who are trying to impersonate authorized personnel to gain access to sensitive information.
Current state & the Future for Behavioral Biometrics
The behavioral biometrics market revenue totaled ~US$ 1.1 Bn in 2020, according to Future Market Insights (FMI). The overall market is expected to reach ~US$ 11.2 Bn by 2031, growing at a CAGR of 23.6% for 2021 – 31. (Future Market Insights, 2021)
As we can see, an increasing number of institutes, financial companies, website owners are using behavioral biometrics in their systems to detect fraudulent usage. The Royal Bank of Scotland uses it to monitor visitors to their websites and apps, others use it in their applications to monitor and authenticate users as an extra verification layer. (PYMNTS.com, 2018)
With the increase in processing capacity, sensor sensitivity and processing algorithms systems can make more accurate identifications of individual users. This allows systems to detect bots, password sharing/compromise.
Ecommerce sites have increasingly started incorporating this technology into their setup to prevent fraud. It can also potentially allow systems to make an educated judgment about the visitor’s gender and age to show appropriate products.
Considering the advantages and minimal hardware investment we will only see an increase in the use of Behavioral Biometrics for authentication in the future.
Advantages of using Behavioral Biometrics for authentication
Behavioral Biometrics have the following advantages that make them attractive for companies and institutes to implement:
- Flexibility: The data being analyzed is not limited to currently identified sets that we have discussed so far. Since most of the processing being done is on the software side the organization can easily add additional behavioral data to be analyzed and processed.
- Convenience: This a major plus point for the technology is that it is a passive layer of security. This allows it to work without interfering with the user workflows. This removes a major obstacle in incorporating security into the system as the traditional security setups decrease the usability of the system.
- Efficiency: They can be applied in real-time to detect fraudulent use and the system can be run against historic data as well to detect improper use after fact.
- Security: Behavioral characteristics are hard to replicate and thus incorporating this additional layer of security improves the security of the system.
Disadvantages of using Behavioral Biometrics for authentication
As with all systems there are some disadvantages of using a Behavioral Biometric system for authentication as well. If we are using the Keystroke analysis then the text being entered has to be long enough for the system to generate a profile and match it so if we are only using it as an additional validation step during password entry and the user’s password is too short, then the system might not be able to create and match a profile.
Another problem is that a user’s behavior can change drastically due to various valid reasons and that can cause access issues when the algorithm is unable to account for the changes. Some of the reasons can include:
- Illness or Injury: If a person is injured or unwell then their usage patternswill change
- Stress
- Pregnancy
- Sleep deficiency
- Caffeine deficiency or overindulgence
- Tiredness: If a user logs back in after a session in the gym their usage patterns are going to differ from the pattern before their gym session
- Time of day: Some people are more active during certain times of day so their usage patterns will vary based on the time of the day.
- Distractions: If the user is distracted while working , or example, if they are on a call and working at the same time. Their behavior patterns will be different.
- Location: If the person logs in from a different location and are working with a different setup their metrics are going to be different. For example the profile when using an egronomic keyboard in office vs using a laptop keyboard while working remotely will be drasticly different and the system will have a hard time creating a consolidated profile for such users.
Another major issue with this technology is the Privacy implications. If we are implementing a system that monitors every keystroke and mouse movement and logs it for analysis then that has a serious privacy implication as sensitive data that shouldn’t be logged such as medical information, personal account passwords, other sensitive information etc can get logged as well. Once the data is logged there is a possibility of data leaks or a breach of the security system which would expose the collected information to an attacker.
Depending on the user’s location collection of this kind of data can be illegal due to rules such as the GDPR (Krausová, 2018), the California Consumer Privacy Act (CCPA) and other such rules. They will also limit the information that can be transmitted across state & country boundaries which can be a concern for multinational companies.
Finally incorporating the processing required for behavior analysis on the local system can be resource intensive which might make the setup infeasible for older machines. If the processing of the data is consolidated at a central location then the usage data would need to be transmitted to the location over the network that can potentially max out the bandwidth and depending on network congestion cause unacceptable delays in the processing and access.
Results and Recommendations
Based on our review of the current state of Behavioral Biometrics in the industry and the technological state of the system/algorithms we find that the technology does help increase the security of the system by adding an additional layer of security to the system. However, it is not yet mature enough to deploy for general commercial implementation and should only be used for securing highly sensitive systems and infrastructure where the security considerations outweigh the limitations identified earlier in the paper.
Once the technology is more mature and the issues identified earlier have been mitigated it can slowly be incorporated in the general computing world as an optional additional layer of security. At no point should this be used as the only layer of security for any system.
Conclusion
Behavioral Biometrics as a security measure is a technology still in its early stages of use and implementation and while it does add an additional layer of security the current limitations do not justify a general release and implementation in general use computing. The system should only be implemented in systems such as classified military systems, critical corporate servers containing highly sensitive information etc where the benefits or security concerns outweigh the disadvantages of using a technology that still needs to mature more.
References
Alzubaidi, A., & Kalita, J. (2016). Authentication of smartphone users using behavioral biometrics. IEEE Communications Surveys & Tutorials, 18(3), 1998–2026. https://doi.org/10.1109/comst.2016.2537748
Araujo, L. C. F., Sucupira, L. H. R., Lizarraga, M. G., Ling, L. L., & Yabu-Uti, J. B. T. (2005). User authentication through typing biometrics features. IEEE Transactions on Signal Processing, 53(2), 851–855. https://doi.org/10.1109/tsp.2004.839903
Banerjee, S. P., & Woodard, D. (2012). Biometric authentication and identification using Keystroke Dynamics: A survey. Journal of Pattern Recognition Research, 7(1), 116–139. https://doi.org/10.13176/11.427
Barral, C., & Tria, A. (2009). Fake fingers in fingerprint recognition: Glycerin supersedes gelatin. Formal to Practical Security, 57–69. https://doi.org/10.1007/978-3-642-02002-5_4
Benchoff, B. (2016, January 18). Emulating and cloning smart cards. Hackaday. Retrieved June 27, 2022, from https://hackaday.com/2016/01/18/emulating-and-cloning-smart-cards/
Bo, C., Zhang, L., Li, X.-Y., Huang, Q., & Wang, Y. (2013). Silentsense. Proceedings of the 19th Annual International Conference on Mobile Computing & Networking – MobiCom ’13. https://doi.org/10.1145/2500423.2504572
Das, R. (2020, October 14). A behavioral biometric – keystroke recognition. A Behavioral Biometric – Keystroke Recognition. https://resources.infosecinstitute.com/topic/a-behavioral-biometric-keystroke-recognition/
Future Market Insights. (2021, October). Behavioral biometrics market. Future Market Insights. https://www.futuremarketinsights.com/reports/behavioral-biometrics-market
Goel, V., & Perlroth, N. (2016, December 14). Yahoo says 1 billion user accounts were hacked. The New York Times. https://www.nytimes.com/2016/12/14/technology/yahoo-hack.html
Krausová, A. (2018). Online behavior recognition: Can we consider it biometric data under GDPR? Masaryk University Journal of Law and Technology, 12(2), 161–178. https://doi.org/10.5817/mujlt2018-2-3
Morris, C. (2021, June 30). LinkedIn data theft exposes personal information of 700 million people. Fortune. https://fortune.com/2021/06/30/linkedin-data-theft-700-million-users-personal-information-cybersecurity/
PYMNTS.com. (2018, August 15). What’s behind the rise of behavioral biometrics? PYMNTS.com. Retrieved June 27, 2022, from https://www.pymnts.com/fraud-prevention/2018/behavioral-biometrics-uk-banks-authentication-security-privacy/
Shmelkin, R., Friedlander, T., & Wolf, L. (2021). Generating master faces for dictionary attacks with a network-assisted Latent Space evolution. 2021 16th IEEE International Conference on Automatic Face and Gesture Recognition (FG 2021). https://doi.org/10.1109/fg52635.2021.9666968
Workos. (2020, September 5). A developer’s history of authentication – WorkOS. A Developer’s History of Authentication. https://workos.com/blog/a-developers-history-of-authentication
Note: This was originally written as a paper for one of my classes at EC-Council University in Q2 2022.
– Suramya
