1. 1 Attribute-Based Encryption Brent Waters SRI International

2. 2 Server Mediated Access Control Access list: John, Beth, Sue, Bob Attributes: “Computer Science”, “Admissions” File 1 Server stores data in clear Expressive access controls

3. 3 Distributed Storage Scalability Reliability Downside: Increased vulnerability

4. 4 Traditional Encrypted Filesystem File 1 Owner: John File 2 Owner: Tim  Encrypted Files stored on Untrusted Server  Every user can decrypt its own files  Files to be shared across different users? Credentials? Lost expressivity of trusted server approach!

5. 5 A New Approach to Encrypting Data File 1 “Creator: John” “Computer Science” “Admissions” “Date: 04-11-06” File 2 “Creator: Tim” “History” “Admissions” “Date: 03-20-05”  Label files with attributes Goal: Encryption with Expressive Access Control

6. 6 File 1 “Creator: John” “Computer Science” “Admissions” “Date: 04-11-06” File 2 “Creator: Tim” “History” “Admissions” “Date: 03-20-05” Univ. Key Authority OR AND “Computer Science” “Admissions” “Bob” A New Approach to Encrypting Files

7. 7 Attribute-Based Encryption [Sahai-Waters 05]  Start with monotonic access formulas [GPSW06]  Techniques from IBE [S84,BF01]  Challenge: Collusion Resistance  Further developments of ABE  Bringing into Practice

8. 8 Attribute-Based Encryption  Ciphertext has set of attributes  Keys reflect a tree access structure  Decrypt iff attributes from CT satisfy key’s policy OR AND “Computer Science” “Admissions” “Bob” “Creator: John” “Computer Science” “Admissions” “Date: 04-11-06”

9. 9 Central goal: Prevent Collusions  If neither user can decrypt a CT, then they can’t together AND “Computer Science” “Admissions” AND “History” “Hiring” Ciphertext = M, {“Computer Science”, “Hiring”}

10. 10 A Misguided Approach K History, K CS, K Hiring, K Admissions, … Public Parameters SK CS, SK Admissions SK History, SK Hiring CT= E K CS ( R), E K Hiring (M-R) Neither can decrypt alone, but …

11. 11 Our Approach Two key ideas  Prevent collusion attacks  Bilinear maps “tie” key components together  Support access formulas  General Secret Sharing Schemes

12. 12 Bilinear Maps  G, G T : multiplicative of prime order p.  Def: An admissible bilinear map e: G  G  G T is: –Non-degenerate: g generates G  e(g,g) generates G T. –Bilinear: e(g a, g b ) = e(g,g) ab  a,b  Z, g  G –Efficiently computable. –Exist based on Elliptic-Curve Cryptography

13. 13 Secret Sharing [Ben86]  Secret Sharing for tree-structure of AND + OR OR AND “Computer Science” “Admissions” “Bob” y y y r (y-r) Replicate secret for OR’s. Split secrets for AND’s.

14. 14 The Fixed Attributes System: System Setup Public Parameters g t 1, g t 2,.... g t n, e(g,g) y “Bob”, “John”, …, “Admissions” List of all possible attributes:

15. 15 Encryption Public Parameters g t 1, g t 2, g t 3,.... g t n, e(g,g) y Ciphertext g st 2, g st 3, g st n, e(g,g) sy Select set of attributes, raise them to random s M File 1 “Creator: John” (attribute 2) “Computer Science” (attribute 3) “Admissions” (attribute n)

16. 16 Key Generation Public Parameters Private Key g y 1 /t 1, g y 3 /t 3, g y n /t n g t 1, g t 2,.... g t n, e(g,g) y Fresh randomness used for each key generated! Ciphertext g st 2, g st 3, g st n, e(g,g) sy M OR AND “Computer Science” “Admissions” “Bob” y y y r (y-r) y3=y3= yn=yn= y1=y1=

17. 17 Decryption e(g,g) sy 3 e(g,g) sy n = e(g,g) s(y-r+r) = e(g,g) sy (Linear operation in exponent to reconstruct e(g,g) sy ) Ciphertext g st 2, g st 3, g st n, Me(g,g) sy Private Key g y 1 /t 1, g y 3 /t 3, g y n /t n e(g,g) sy 3

18. 18 Security  Reduction: Bilinear Decisional Diffie-Hellman  Given g a,g b,g c distinguish e(g,g) abc from random  Collusion resistance  Can’t combine private key components

19. 19 The Large Universe Construction: Key Idea Public Function T(.), e(g,g) y Private Key  Any string can be a valid attribute Ciphertext g s, e(g,g) sy M For each attribute i: T(i) s For each attribute i g y i T(i) r i, g r i e(g,g) sy i Public Parameters

20. 20 Delegation AND “Computer Science” “admissions” OR “ Bob ”  Derive a key for a more restrictive policy Year=2006 Bob’s Assistant

21. 21 Making ABE more expressive  Any access formulas Challenge: Decryptor ignores an attribute  Attributes describe CT, policy in key Flip things around

22. 22 Supporting “NOTs” [OSW07] Example Peer Review of Other Depts. AND “Year:2007” “Dept. Review” “Computer Science” NOT Bob is in C.S. dept => Avoid Conflict of Interest Challenge: Can’t attacker just ignore CT components?

23. 23 A Simple Solution  Use explicit “not” attributes  Attribute “Not:Admissions”, “Not:Biology”  Problems: Encryptor does not know all attributes to negate Huge number of attributes per CT “Creator: John” “History” “Admissions” “Date: 04-11-06” “Not:Anthropology” “Not:Aeronautics” … “Not:Zoology”

24. 24 Technique 1: Simplify Formulas Use DeMorgan’s law to propagate NOTs to just the attributes AND “Dept. Review” “Public Policy” “Computer Science” NOT OR NOT

25. 25 Applying Revocation Techniques  Broadcast a ciphertext to all but a certain set of users  Used in digital content protection E.g. Revoke compromised players P1P1 P2P2 P3P3

26. 26 Applying Revocation Techniques  Focus on a particular Not Attribute AND “Year:2007” “Dept. Review” “Computer Science” NOT

27. 27 Applying Revocation Techniques  Focus on a particular ‘Not’ Attribute “Computer Science” NOT “Creator: John” “Computer Science” “Admissions” “Date: 04-11-06”  Attribute in ‘Not’ as node’s “identity”  Attributes in CT as Revoked Users Node ID not in “revoked” list =>satisfied N.B. – Just one node in larger policy

28. 28 The Naor-Pinkas Scheme  Pick a degree n polynomial q( ), q(0)=a n+1 points to interpolate  User t gets q(t)  Encryption: g s,,Mg sa Revoked x 1, …, x n g sq(t) g sq(x 1 ),..., g sq(x n ) Can interpolate to g sq(0) =g sa iff t not in {x 1,…x n }

29. 29 Applying Revocation to ABE  Use same S.S. techniques for key generation Same techniques for pos. attributes  “Local” N-P Revocation at each Not-Attribute  Upshot: N-P Revocation requires to use each CT attribute

30. 30 Ciphertext Policy ABE [BSW07]  Encrypt Data reflect Decryption Policies  Users’ Private Keys are descriptive attributes OR AND “Discipline Committee” “Professor” “Counselor” “Professor”, “Discipline Committee”, “Age=33”, “History” Univ. Key Authority “Thinking” Encryptor

31. 31 Challenges in Practice [PTMW06]  Applications Health Care Netflow Logs (currently building)  How are CTs annotated? Can we automate?  Convention for using Attributes? “Prof.” or “Professor” Does “T.A.” + “CS236” mean TAing CS236?

32. 32 Challenges in Practice  What group do Public Parameters represent? Univ. Key Authority Individual’s Key

33. 33 Advanced Crypto Software Collection  Goal: Make advanced Crypto available to systems researchers  http://acsc.csl.sri.com (8 projects) http://acsc.csl.sri.com $ cpabe-setup $ cpabe-keygen -o sara_priv_key pub_key master_key \ sysadmin it_department 'office = 1431' 'hire_date = '`date +%s` $ cpabe-enc pub_key security_report.pdf (sysadmin and (hire_date = 5, audit_group, strategy_team)) Projects at UIUC and MIT using ABE

34. 34 Conclusions and Open Directions  Attribute-Based Encryption for Expressive Access Control on Encrypted Data  Extending Capabilities Delegation Non-Monotonic Formulas Ciphertext-Policy  Currently implemented

35. 35 Conclusions and Open Directions  Open: Can we express access control for any circuit over attributes?  What are limits of capability-based crypto? Capability that evaluates any function s Univ. Key Authority F( ) F(s)

36. 36 Thank You

37. 37 Related Work  Identity-Based Encryption [Shamir84,BF01,C01]  Access Control [Smart03], Hidden Credentials [Holt et al. 03-04] Not Collusion Resistant  Secret Sharing Schemes [Shamir79, Benaloh86…] Allow Collusion

38. 38 System Sketch Public Parameters Choose degree n polynomial q(), q(0)=b Can compute g q(x) g q(0), g q(1),.... g q(n), Ciphertext g s, g sq(x 1 ), …, g sq(x n ) Attributes: x 1, x 2 … =t Private Key g rq(t), g r “Computer Science” NOT e(g,g) srq(t) e(g,g) srq(x 1 ) e(g,g) srq(x n ) If points different can compute e(g,g) srb

39. 39 Applications: Targeted Broadcast Encryption  Encrypted stream AND “Soccer” “Germany” AND “Sport” “11-01-2006” Ciphertext = S, {“Sport”, “Soccer”, “Germany”, “France”, “11-01-2006”}

40. 40 Extensions  Building from any linear secret sharing scheme  In particular, tree of threshold gates…  Delegation of Private Keys

41. 41 Threshold Attribute-Based Enc. [SW05]  Sahai-Waters introduced ABE, but only for “threshold policies”: Ciphertext has set of attributes User has set of attributes If more than k attributes match, then User can decrypt.  Main Application- Biometrics

42. 42 Central goal: Prevent Collusions  Users shouldn’t be able to collude AND “Computer Science” “Admissions” AND “History” “Hiring” Ciphertext = M, {“Computer Science”, “Hiring”}