HumanAUT Secure Human Identification Protocols

1
HumanAUTSecure Human Identification Protocols

• Adam Bender
• Avrim Blum
• Manuel Blum
• Nick Hopper

The NSF ALADDIN Project Carnegie Mellon University
2
The HumanAUT mini-PROBE

• Part of the Computer-Human Authentication PROBE
• HumanAUT stands for Human AUThentication
• Authentication proving your unique identity
  (logging in) to a third party
• Focus is on designing protocols for a human to
  use for authentication

3
What kind of protocols?

• HumanAUT is a challenge-response protocol using a
  shared secret between a computer, which generates
  a random challenge on demand, and a human, who
  answers these challenges using the shared secret
• The correct response depends on the shared secret
  which only the authorized human and computer
  know

4
Motivation flaws in current protocols in
applications (1)

• Passwords can be snooped, key-logged, sniffed,
  cracked, guessed
• PIN numbers are short and often easy to guess
  (birthdays), susceptible to shoulder surfing
• Traditional challenge-response has small set of
  fixed responses with easily obtainable answers

5
Motivation flaws (2)

• Hardware is expensive and relies on physical
  mechanisms, which can be stolen or lost
• Biometrics are also expensive, and not as secure
  as we thought
• Gelatin fingers (Matsumoto 02)
• Able to reconstruct sample images from face
  recognition template (Adler 03)

6
The HumanAUT environment

• Assume you are a
• Naked person
• In a glass house
• With an insecure terminal
• Using a short secret over and over
• Or you have lost your luggage, or had your wallet
  stolen, etc.
• How do you authenticate yourself in the presence
  of adversaries?

7
Necessary properties of authentication protocols

• Secure
• No one else can authenticate themselves, even
  after observing successful authentications
• Human executable
• People have to do it in their heads without
  hardware or other aids anything that does not
  directly involve their brain can be forged

8
Theory to the rescue

• Can provide schemes that are computationally
  secure
• Even when adversary is watching
• Current schemes are based on problems that are
  hard on average
• Attempt to make efficient tradeoff with human
  executability

9
Current work

• Based on what are suspected to be hard machine
  learning problems
• Draws from areas of security, machine learning,
  complexity theory, algorithms has an impact on
  security software (anything that requires
  authentication), banking

10
Toy example
Challenge is a set of digits on a map
2
6
7
4
8
9
4
3
0
6
1
4
2
3
5
1
11
Toy example
Challenge is a set of digits on a map Secret is
a predetermined subset of locations
2
6
7
4
8
9
4
3
0
6
1
4
2
3
5
1
12
Toy example
Challenge is a set of digits on a map Secret is
a predetermined subset of locations And a parity
digit location
2
6
7
4
8
9
4
3
0
6
1
4
2
3
5
1
13
Toy example
If the parity digit is even, response is the sum
mod 10 of the secret locations 405 9
2
6
7
4
8
9
4
3
0
6
1
4
2
3
5
1
14
Toy example
If the parity digit is even, response is the sum
mod 10 of the secret locations 405 9 If it
is odd, response is the sum plus the
parity digit 4905 8 (mod 10)
2
6
7
4
9
9
4
3
0
6
1
4
2
3
5
1
15
Long-term security

• Given a large number of such challenges and their
  responses, it should be hard to determine the
  secret locations
• Thus there is a negligible probability of an
  adversary successfully authenticating himself

16
Sample
6
4
1
1
0
9
2
9
7
2
8
5
4
7
2
3
17
Current focus

• Design a better scheme
• Based on something that is easy for people to do
• Prove a strong security result
• Implement this scheme
• Create a demo to collect data on how easy this is
  for people to use
• Challenge anyone to break it