7. Fingerprints on Device
Just asking to be broken:
• Insecure storage on device
Insecure storage in cloud
• On-device enclave
easily hacked / not encrypted
8. Basic Exploit that actually works
(on some Android phones)
• Asdf
• Etched PCB & Alumninum Foil (Starbug)
• asdf
9. How to Hack Fingerprints
• Asdf
• Etched PCB & Alumninum Foil (Starbug)
• asdf
10. Update on Fingerprints
The Big Exploit (2018)
• Deep Master Print – Philip Bontrager
& Academic Team at NYU
• A machine learning driven exploit that
analyzed a number of fingerprints in
order to build a 3D model fingerprint
that matches a large portion of fingers
used on for secure login on devices today.
11. Facial Recognition Exploits
• Facial scans work by matching characteristics of a face
to a template enrolled in a DB.
Basic “blocks” on face recognizers are known:
• Adding obfuscation and visual confusion
• Even wearing a hat and sunglasses can muck up a facial
scan
• Downside of most facial “obfuscation” hacks is that it can
be recognized by other human beings
More advanced exploits to fake the results:
• Machine learning derived fake faces
• AI-driven creation of face from multiple angles
• 3D printing of 3D faces, with fake liveliness
(hard to do, but academics have proven it’s doable)
13. Evolution of Facial Recog Exploits *
* Original work by Yi Xu, True Price, Jan-Michael Frahm, and Fabian Monrose
Department of Computer Science, University of North Carolina at Chapel Hill USENIX Security
14. Evolution of Facial Recog Exploits *
* Original work by Yi Xu, True Price, Jan-Michael Frahm, and Fabian Monrose
Department of Computer Science, University of North Carolina at Chapel Hill USENIX Security
15. Evolution of Facial Recog Exploits *
* Original work by Yi Xu, True Price, Jan-Michael Frahm, and Fabian Monrose
Department of Computer Science, University of North Carolina at Chapel Hill USENIX Security
16. How to Fake a Facial Scan: 3D Heads
• Reproduction of Facial Recog Areas only (higher fidelity)_
17. Iris Scan Exploits
• Iris scans appear to be highly
secure, because it is scanning a
unique body part under high
resolution.
However, it can be hacked:
• Contact Lens can fake an iris
• Upload of a infrared scan of a
person’s face (no access to
reference data, instead, just an
infrared scan of a eye at high rez)
• Requires technical expertise
• Newer hacks require a scan of the
iris – hack of reference data
18. Iris Scan Exploits
• Examples:
Eye spy
By Chaim Gartenberg @cgartenberg May 23, 2017, 10:37am EDT
TECH SAMSUNG CYBERSECURITY
Hacker beats Galaxy S8 iris scanner using an IR
image and a contact lens
11
Based on name alone, the futuristic iris-scanning feature on the Galaxy S8 sounds like it
would be the most secure way to lock your phone. Hacker Jan Krissler, who goes by the
name Starbug, shows in a recent video that, despite the impressive technology in
unlocking your phone with your eyes, the security system can be beaten with a relatively
low-tech hack.
As the video shows, Starbug is able to take a infrared picture of a person’s face using
the night mode setting on a regular point and shoot camera. Print it out on an ordinary
laser printer and it fools the camera by placing a contact lens over the image to give it
the appearance of an actual human eye. While it certainly is a little more effort than, say,
(https://1.bp.blogspot.com/-rSiTjwXZmT4/VPmbURLovxI/AAAAAAAAiH0/jB3L24BeGO0/s728-
e100/iris-biometric-security-system.jpg)
Hacker Finds a Simple Way to Fool IRIS Biometric
Security Systems
March 06, 2015 Swati Khandelwal
Biometric security systems that involve person's unique identi cation (ID), such as
Retinal, IRIS, Fingerprint or DNA, are still evolving to change our lives for the better
even though the biometric scanning technology still has many concerns such as
information privacy, and physical privacy.
In past years, Fingerprint security system (https://thehackernews.com/2013/09/ nally-
iphones- ngerprint-scanner.html) , which is widely used in different applications such as
smartphones and judicial systems to record users' information and verify person's
identity, were bypassed several times by various security researches, and now, IRIS
scanner claimed to be defeated.
19. Veins / Palm Exploits
• Vein / Palm scans
were thought to be
highly secure alternative
to fingerprints
• Turns out that these
can be hacked as well
(with reference data)
22. Biometric Identity Processing System
• Input Data (1)
• Input Data (1a)
• Reference Data (1b)
• Sensor (2)
• Software (3)
• Matcher
• Threshold
Sensor
Software
Preprocessing
Matching
Database
(2)
(3)
(1b)(1a)
Input
Data
(1a)
Structure of this system originally outlined in this format by Starbug, 2014
23. 3 Types of Attacks
Sensor
Software
Preprocessing
Matching
Database
(2)
(3)
(1b)(1a)
Input
Data
(1a)• Attack the Input Data (1)
• Input Data (1a)
• Reference Data (1b)
• Attack using the Sensor (2)
• Attack the Software (3)
• Matcher
• Threshold
Sensor
Software
Preprocessing
Matching
Database
24. 1. Attack Via Input Data
• Attack the Input Data (1)
• Input Data (1a)
• Most Common Attack Vector:
Easiest and most accessible vulnerability
• Reference Data (1b)
• No Attacks recently directly along this vector
• But high-fidelity hacks require access
to cracked original Reference data
Sensor
Database
(1b )(1a)
Software
Input
Data
Reference
Data
(1a)
25. 2. Attack Via Sensor
• Attack the Input Data (1)
• Input Data (1a)
• Reference Data (1b)
• Attack using the Sensor (2)
Sensor
Software
Preprocessing
Database
(2)
(1b)(1a)
Input
Data
(1a)
26. 2. Attack Via Software
• Attack the Input Data (1)
• Input Data (1a)
• Reference Data (1b)
• Attack using the Sensor (2)
• Attack the Software (3)
• Matcher
• Threshold
Sensor
Software
Preprocessing
Matching
Database
(2)
(3)
(1b)(1a)
Input
Data
(1a)
28. Multi-factor authentication
• NIRVANA: Multiple biometrics + Identity Face match / PIV-I card check
validation by an in-person check with actual human (military grade)
• BETTER FOR BUSINESS: Multi-factor authentication which includes
but does not privilege biometrics – treats data knowledge as equivalent
• Multiple biometrics + PIN/Login / Passcode
• PRETTY GOOD SECURITY: Multi-factor biometric security which
occurs simultaneously (pretty hard to hack all in sync)
• Fingerprints + Facial Recognition + Iris + Audio Recognition
• Note: Requires enrollment/login stations capable of handling multiple biometrics
BEST
BETTER
GOOD
29. High fidelity / Multi-finger enrollment
• Most fingerprint systems (on device) only collect and store a few
millimeters of a fingertip.
• This small sample set is relatively easy to replicate and use in a hack.
• To prevent this hack, use a higher fidelity enrollment system that
enrolls more area of the finger and more fingers on each hand.
VS.
Collect much more data,
match on many more points
30. Facial Recognition
• Facial recognition systems also operate off a limited template
• Adding complexity to the input is useful - ensure you are
capturing not only the front face, but also the side, the back, as
much movement as possible
• Add Liveliness detection + multi-angles
• Collect much more data,
match on many more points
VS.
31. Software
How to Prevent 3 Types of Attacks
• Complicate/Harden the Input Data (1)
• Provide Observation of Sensor (2)
• Harden the Software (3)
Preprocessing
Matching
Database
(2)
(1b)(1a)
Input
Data
(1a)
(3)
Sensor
32. 1. Harden/Complicate Input Data
• Complicate/Harden the Input Data (1)
• Input Data (1a)
Database
(1a)
Input
Data
(1a)
Sensor
Software
33. 1. Harden/Complicate Input Data
• Complicate/Harden the Input Data (1)
• Input Data (1a)
• Add multiple biometrics that login
simultaneously (not sequentially)
• Require higher fidelity enrollment
and more data from each biometric
• Add more minutiae as input data
Database
(1b)(1a)
Input
Data
(1a)
Input
Data
+
Sensor
Software
34. 2. Add Observation of Sensor
Database
(2)
(1b)(1a)
Input
Data
(1a)• Complicate/Harden the Input Data (1)
• Provide Observation of Sensor (2)
• IDEAL – IN PERSON: Have an actual person
observe both enrollments and login
(this can be done remotely & off-shore)
• RANDOM SCREENS: Randomly
audit logins with human observation
• AI OBSERVATION: Add layer of
observational video and AI to check
humans at the enrollment station
and actions at station. Check multiple
signifiers of actual human activity
(voice, movement, approach to station, etc.)
Sensor
Software
35. 2. Add Observation of Sensor
• Complicate/Harden the Input Data (1)
• Provide Observation of Sensor (2)
• IDEAL – IN PERSON: Have an actual person
observe both enrollments and login
(this can be done remotely & off-shore)
• RANDOM SCREENS: Randomly
audit logins with human observation
• AI OBSERVATION: Add layer of
observational video and AI to check
humans at the enrollment station
and actions at station. Check multiple
signifiers of actual human activity
(voice, movement, approach to station, etc.)
Sensor
Database
(2)
(1b)(1a)
Input
Data
(1a)
Software
36. Software
3. Harden the Software
Sensor
Preprocessing
Matching
Database
(2)
(3)
(1b)(1a)
Input
Data
(1a)
• Complicate/Harden the Input Data (1)
• Communication Data (1a)
• Reference Data (1b)
• Provide Observation of Sensor (2)
• Harden the Software (3)
• THRESHOLD: ideal to raise threshold
to accommodate high fidelity logins
(adds enrollment and login time obviating
some reasons to use biometrics in the first place)
• PROCESSING: use hardened pre-processing
with templates that provide encrypted
matching algorithms / store templates securely
• MULTI-FACTOR MATCHING: Match against multiple
biometrics simultaneously, not just one input at a time.
37. Software
3. Harden the Software
• Complicate/Harden the Input Data (1)
• Communication Data (1a)
• Reference Data (1b)
• Provide Observation of Sensor (2)
• Harden the Software (3)
• THRESHOLD: ideal to raise threshold
to accommodate high fidelity logins
(adds enrollment and login time obviating
some reasons to use biometrics in the first place)
• PROCESSING: use hardened pre-processing
with templates that provide encrypted
matching algorithms / store templates securely
• MULTI-FACTOR MATCHING: Match against multiple
biometrics simultaneously, not just one input at a time.
Sensor
Preprocessing
Database
(2)
(1b)(1a)
Input
Data
(1a)
MatchingMatchingMatchingMatching
(3)
38. Software
A Hardened Biometrics System
More complicated, but much more secure
• Complicate/Harden the Input Data (1)
• Includes multiple bio inputs
• Enroll at higher fidelity / more minutiae
• Provide Observation of Sensor (2)
• Includes observational data
(actual human ideal)
• Harden the Software (3)
• Higher threshold for enrollment/login
• Includes encrypted template DB
• Includes multi-factor matching
Sensor
Preprocessing
Matching
Database
(2)
(1b)
(1a)
Input
Data
(1a)
(3)
MatchingMatchingMatching