Privacy preserving telemetry

1. Privacy Preserving Telemetry in Reality, an
illusion bound to silicon?
A.Nappa - F. Olteanu - M. Dhiman
1

2. A Long Story
Can a human controlled computer make itself-himself distinguishable from an automated program?

3. The Multiple Facets of Trust
3

4. TRUST
zkCrypto
Performance
Scalability
Resiliency
Software
Side Channels - Architectures - Physics
Human Computer Interaction
Differential Privacy
Crowd-Blending Privacy
Human
Attestation
Privacy
Trusted
Hardware
Quantum
VPN0
AD-Injection

5. 5
TRUST
zkCrypto
Performance
Scalability
Resiliency
Software
Side Channels - Architectures - Physics
HCI
Differential Privacy
Crowd-Blending Privacy
Human
Attestation
Privacy
Trusted
Hardware
Quantum
VPN0

6. Scope of this talk:

7. Motivation
7

8. Popular Option: Google reCAPTCHA
● Bad user experience
● With ML, bots may do better than humans
● Privacy concerns
● Repeatedly Broken
8

9. 9
Tap
Events
Simulated
Tap
Events
Acc
Outputs
Acc
Outputs
Bot Detection

10. Maintaining the privacy of the user
1
0

11. What are the issues
1
1
SHM
CPU
SENSOR

12. Recipe
12
● GPS data
● Geohash
● BTS data
● SafetyNet (Android)
● DeviceCheck (Apple)
● ZoKrates

13. Geohash:
13
A: gcpvn5bpncpy
B: gcpvjfz795xc0
A gcpvn5bpncpy
B gcpvjfz795xc0

14. Approach: Socrates House
14

15. Approach: ZoKrates House
15

16. 16
GPS Coordinates
Time Stamp
Hashed Values
Hashed Values
PoK pre-image
ZKP pre-image
validates certain
constraints
“I prove that I
have been in
two different
points (signed
by a trusted
entity)
separated by a
distance greater
than a threshold
and in a given
range of time”
Proving in Zero-Knowledge

17. 17
Proving in Zero-Knowledge

18. Robustness
Assumption 1 (ideal world)
● The BST sends signed data that can be verified and
triangulated with GPS to protect GPS spoofing
[Oligieri et al. ACM WiSec19]
18
Assumption 2 (Our world)
● Interleave Safetynet/Devicecheck attestations with
GPS and BST data, triangulate, attest again

19. Attacks
19

20. Meet OpenTitan
20

21. Implementation
2
1
PMP
SHM
RISCV
CPU
SENSOR SK
SK

22. Now we have hi-fi data from sensors
2
2

23. Evaluation
Number of gates for path range: 444K
Prover time (1 distances): ~16s
Total proof (linking two generated proofs): 20s (on PC)
Proof size: ~5MB
23

24. Choose your battle:
● Tradeoff #users #transactions vs. performance
● How often we proof?
● Which sensors are most fiable?
● Which are the weakest?
● Combining them how much battery drains? Plus the proof? How many times
a day?
● CAN PROOF BE FORGED?

25. Conclusion
● Our battle is to provide trusted data from sensors (i.e. blockchain oracles)
● Many users will be involved
● We will need an efficient solution, probably using textbook RSA signatures

26. The Attack of the Clones: A Study of the Impact of
Shared Code on Vulnerability Patching

27. We need to be in the machine
WINE-BR
► 8 Million Windows Hosts
Internet Users
Symantec Users
WINE Opted-in
WINE Users
10 Apps

28. One raɣ to rule them all: Inducing soft-errors by
gamma-ray emissions at ground level
28

29. Motivation
29

30. Cosmic Rays
30
I
I
I

31. Ground Level: Belgian Election Vs. Apple Vs. FBI
31
x …. 0 0 0 0 0
1 1 1 1 0
0 1
0
1 = 6080
= 1984

32. Can we reach an exploit from a neutron?
32
A system on Earth, at sea
level, with 4GB of RAM has
a 96% percent chance of
having a bit error in three
days without ECC RAM.