I Know your PIN I Know Your PIN Jolyon Clulow Prism jolyonc@prism.co.za www.prism.co.za

What this talk is not about: What this talk is about: Introduction What this talk is not about: The Internet, SSL, VPNs What this talk is about: Bank, Credit and Debit cards Banks, Financial Networks and Switches PINs, PANs, ATMs, TRSMs, POS, Mobile And… …(some of the) ways that one can recover PINs from such supposedly secure systems!

So why are we interested? Justification Driver for modern cryptography – the so called ‘killer app’ of cryptography The concept of a ‘PIN’ is internationally understood and accepted Scale of use Bank, Credit and Debit Cards Card issuing banks Card Associations (Visa, MasterCard,etc) Amount of money protected by these operations Guaranteed that (almost) everyone who reads this, relies on the security thereof to protect their own personal finances.

Talk Outline Introduction Background info: What is PIN security? The attacks Some remedies Real world scenarios The road ahead? Conclusion

An Introduction to PINS Background info Financial Security 101: An Introduction to PINS

Terminology PIN: Personal Identification Number PAN: Personal Account Number ATM: Automatic Teller Machine (cash machine) API: Application Programming Interface (the set of functions exposed/available) API attack: an attack which uses(or abuses) the existing/available functions to compromise the security of the system

What is a TRSM? Tamper Resistant/Responding Security Module (TRSM) Host Security Module (HSM) Hardware Security Module (HSM) Crypto Coprocessor Provides a secure, trusted environment to perform sensitive operations Detects and responds to physical, electronic (or other) attempts to recover key material or sensitive data. Typical measures include: physical tamper envelope/membrane temperature, radiation sensors power supply monitoring and filtering Trigger causes erasure of protected data

Financial Network Architecture

Key Zones Each connected pair of entities share a common key to form a key zone

Basic Operations 3 Basic PIN operations are required: Encryption Translation Verification

PIN Encryption e.g. PIN is 1234, Key is 0123456789ABCDEF Start with an empty PIN block Insert PIN Pad Encrypt the clear PIN block It’s that simple! 1 2 3 4 1 2 3 4 F 2 5 8 D 6 B 4 9 1

PIN Formats (some examples) VISA Format 3 PIN Block = PPPPFXXXXXXXXXXX IBM 3624 PIN Block = PPPPxxxxxxxxXXXX ISO-1 PIN Block = CLPPPPrrrrrrrrRR where C = X‘1`, L = X‘4` to X’C` r is either P or R VISA Format 2 PIN Block = LPPPPzzDDDDDDDDD

PIN Formats (List). ISO-0 (ANSI X9.8, VISA-1, ECI1) ISO-1 ISO-2 VISA-2, VISA-3, VISA-4 IBM 3624, IBM 3621, IBM 4700 ECI-2, ECI-3 Docutel Others…

ANSI X9.8 Format (ISO-0) E.g. For a 4 digit PIN P1 = 04PPPPFF FFFFFFFF P2 = 0000AAAA AAAAAAAA Where AAAAAAAAAAAA represents 12 digits of the PAN PB = P1  P2 EPB = ek(PB) Binds the account number to the PIN Diversifies the encrypted PIN block

Basic Operations 3 Basic PIN operations are required: Encryption Translation Verification

Translate between different zone keys Question: PIN Translate Translate between different zone keys Question: What if different actors/entities use different formats? Additional operation required PIN Reformat Supports change in PIN formats and PANs

Basic Operations 3 Basic PIN operations are required: Encryption Translation Verification

Exist multiple different approaches PIN Verification Exist multiple different approaches Simple Offsets PIN Verification Values(PVV) Compare the customer supplied PIN with a reference PIN

PIN Verification (Offsets) Validation data is encrypted under PIN generation (verification) key. Ciphertext is ‘decimalised’ to form IPIN by means of a table. Calculate the offset as OFFSET = PIN-IPIN (where ‘-’ is subtraction modulo 10)

PIN Verification (Offsets) IBM PIN Offset Algorithm Allows user to choose own PIN (also to change it easily) Validation data is typically customer and financial institution specific (e.g. PAN) ‘Decimalization’ by means of a table. 1 2 3 4 5 6 7 8 9 A B C D E F

Attack #1a: ANSI X9.8 Attack Attack #1b: Extended ANSI X9.8 Attack The Attacks Attack #1a: ANSI X9.8 Attack Attacks the PIN translate function. Attack #1b: Extended ANSI X9.8 Attack Attacks the PIN translate and reformat functions. Attack #2: The Decimalization Attack Attack against PIN verification algorithm using offsets.

Attack #3: Key Separation #1 The Attacks Attack #3: Key Separation #1 Attack against PIN verification functions based on failure to enforce key separation between verification and translation(encryption). Attack #4: Key Separation #2 Attack against PIN verification functions based on failure to enforce key separation for different verification algorithms. Attack #5: Check Value Attack Attack against PIN verification algorithm using the check value of a key

ANSI X9.8 (ISO-0) Attack Attack #1 Attacks the PIN translate/reformat function

ANSI X9.8 (ISO-0) Attack Input Parameters Attack Strategy: Encrypted PIN Block (EPB) PAN Encrypted ‘In’ Key Encrypted ‘Out’ Key Attack Strategy: In an iterative manner, we make a modification to the PAN and observe the effects

ANSI X9.8 (ISO-0) Attack Under normal operation: Inputs (EPB, P2) PB = dk(EPB) P1 = PB  P2 = 04PPPPFFFFFFFFFF Extract PIN as PPPP Test that PPPP is valid PIN (i.e. each P is a valid decimal digit)

ANSI X9.8 (ISO-0) Attack Instead of supplying the correct PAN (P2) to a call, use a modified PAN (P2’ = P2  ) Inputs (EPB, P2’) PB = dk(EPB) P1’ = PB  P2’ = (P1  P2)  (P2  ) = P1   Say  = 0000x000000000 P1’ = 04PPPPFFFFFFFFFF  0000x000000000

ANSI X9.8 Attack Q: What happens if (P  x) is a decimal digit? A: The call passes. Q: What happens if (P  x) is not a decimal digit? A: Typically, the call FAILS! We have a test for (P  x) < 10.

Building a simple algorithm to identify P ANSI X9.8 Attack Building a simple algorithm to identify P Try all possible values of x, yielding a unique* pattern of ‘passes’ and ‘fails’ allowing you to identify P. A decision tree

The Decimalization Attack Attacks the PIN Verification using offsets function

Decimalization Attack Input Parameters Encrypted PIN Block (EPB) Validation Data Decimalization Table Offset Encrypted Key Attack Strategy: In an iterative manner, we make a single change to an entry in the decimalization table and observe the effects

Decimalization Attack PIN = 6598 PIN Ver Key = 05050505 05050505 Val. Data = 11223344 55667788 Ciphertext = E481FC56 58391418 Dec. Table = 01234567 89012345 IPIN = 4481 Offset = 2117

Decimalization Attack Dec. Table (0) = 11234567 89012345 IPIN = 4481 Offset = 2117 (will pass) Dec. Table (1) = 02234567 89012345 IPIN = 4482 Offset = 2117 (will fail) = 2116 (will pass) Thus far we have identified that the 4th digit in the original IPIN is a 1 and hence that the 4th PIN digit is 1+7 = 8 (IPIN + Offset).

Decimalization Attack Work factor Initial search for (an unknown) offset requires at most 104 + (n-4)•10 queries Each change in the dec. table requires at most 24 + (n-4) queries At most need to try 15 of the 16 entries in the table for a total of 15(24 + n-4) queries. Attack time dependant on TRSM speed Typical values (dependent on speed of TRSM): Known initial offset: 1 – 20 seconds Unknown initial offset: 10 - 1000 seconds

Properties How efficient are these attacks? What are the requirements? Computationally trivial Extremely fast Requires just a few seconds on a Pentium I Typically limited by performance of TRSM What are the requirements? Requires query access to the device, implying either: Physical access to the device/switch/trust center Special case: Stolen device Access to the network transporting transaction traffic and the ability to inject messages

What about in the ‘Real World’? Real world systems should be following standard industry best practices that if implemented correctly and enforced should limit a potential hacker’s ability to perform such attacks. Physical access control to restricted area. Some thoughts and counter arguments. Attacker can attack at weakest point. One institution’s account holder can be compromised on another institution’s network. Hence must guarantee that all potential networks through which the PIN may travel to be secure. So why did you buy an expensive TRSM in the first place if your defense rests on physical access control? Multi-lane Retail Stores

So what went wrong? Some functions are just badly thought out and insecure. Individually secure functions were added to the API in a manner to make entire system insecure. Insufficient attention was given to the possible interplay between functions. Absence of a single standard to which everyone completely adheres to (many different formats and algorithms exist due to historical reasons). Different customers want different functionality from the same product.

Solutions - Cryptographic Remove ‘weaker’ algorithms/functions (leave only the strongest) Parameter(data) Integrity MAC the PIN block and data PAN, PIN block format, etc MAC any verification/generation data Decimalization table, Validation data, TSP, etc A better PIN Block Format? Key Separation Format (PIN Block Variance) Algorithms Other data (e.g. PAN)

Solutions – Access Control Electronic access control Fine grained, allowing the individual enablement/disablement of Formats Algorithms Functions Limit functionality. Only enable what is required. Disable everything else. Useful to allow a function to be disabled should it later be shown insecure. True split control

Risk, Reward and Liability Hackers and Threats? Real world scenarios: Risk, Reward and Liability

Disclaimer This material is made available as a courtesy, purely for educational and informative purposes only for an intended audience of responsible individuals with a genuine interest in improving the security of financial networks. Prism makes no claim as to the accuracy or completeness of this information. Prism accepts no responsibility or liability arising from the use of this material.

Insider attack Extract the PIN number for a given account (or accounts) Create a duplicate ‘white card’ (or multiple duplicate cards) Distribute to accomplices to perform a random tour of ATMs

Insider Attack - Reward Let N be the number of compromised accounts, P the average period before unnatural transaction behavior is noticed and L the daily withdrawal limit. Total Fraud Value = NPL Example: N = 5000 P = 2 L = $1000 Total Fraud = $ 10 M

Account Holder Attack Produce a number of duplicate ‘white cards’ of your own card Distribute to multiple accomplices, preferably in different geographical locations to perform a random tour of ATMs. Report the ‘unauthorized’ activity on your account and dispute the transactions.

Account Holder Attack (cont.) It may be advisable to perform a valid transaction “simultaneously” with a fraudulent one since this ‘proves’ you are in possession of your card and preferably in a different location. Best done by multiple card holders from a given institution since: Not an isolated incident Questions the security of the institution Gives the impression of a possible insider attack

Account Holder Attack - Reward Let N be the number of conspiring account holders, P the average period before unnatural transaction behavior is noticed and L the daily withdrawal limit. Total Fraud Value = NPL Average return = PL Example: N = 100 P = 10 L = $1000 Total Fraud = $ 1 M Average return per account holder = $ 10 K

The Repudiation Attack Just deny a transaction Dispute procedure leading to possible litigation Argue the insecurity of the system Best if security of institution already questioned Scenario: Following a successful account holder/insider attack being made public – other account holders (acting individually) may dispute valid transactions that occurred during the attack period (or after) Financial risk is great due to the possible scale (e.g. 0.1 % of an institution’s 1,000,000 customers each disputing a $1000 transaction = $1 M) Loss of confidence in the given institution could well be more damaging

Other Ideas The Competitor Attack The Stock Market Attack Use own network to compromise a competitor institution (could even choose to use administrator privileges to effect this) Reward not the stolen money but the ‘after effects’ Less of a connection between accomplices and institution (no cash trail leading back) The Stock Market Attack ‘Short’ the stock prior to any attack (no cash trail) The Terrorist Attack All/any combinations of all the previous attacks

What now? Q: What should you do now if you are a bank? Q:Is that all? Contact your vendor, request any best practices information and implement it. Be vigilant. Increase your auditing. Reassure your clients. Wait. Positive pressure on the role players. Q:Is that all? The nature of the problem is such that it is not yours alone (unless you disconnect from the network). The entire network must be secured and until that happens you and your account holders are potentially vulnerable.

The road ahead? Process driven by Card Associations? Due to role and influence over the infrastructure Revise the standards New design/security requirements. Prescriptive requirements limiting what functionality is allowed. Vendors will then update products based on revised standards Expecting (and hoping) for more uniformity and collaboration between different vendor product offerings. (Makes business sense for institutions) Card associations will mandate new requirements to institutions.

The unanswered question? Who is liable in the event of such an attack leading to fraud?

Summary A set of API attacks which allow PIN recovery Design criteria/suggestions to combat the attacks Some potential attack scenarios

The final comment… The most concerning aspect of these attacks, is that you can be attacked on someone else’s network – a network over which you have little or no control.