Towards Robust Fingerprinting of Relational Databases by Mitigating Correlation Attacks.

Area of Science:

• Computer Science
• Data Security
• Cryptography

Background:

• Database fingerprinting is crucial for preventing data leakage and identifying unauthorized sharing.
• Existing fingerprinting schemes are vulnerable to sophisticated correlation attacks exploiting data entry relationships.

Purpose of the Study:

• To identify and demonstrate the effectiveness of correlation attacks (column-wise, row-wise, and integrated) against current database fingerprinting methods.
• To develop and evaluate novel mitigation techniques that enhance the robustness of database fingerprinting against these correlation attacks while preserving data utility.

Main Methods:

• Analysis of existing database fingerprinting vulnerabilities.
• Development of mitigation techniques applicable as post-processing steps.
• Empirical evaluation using a real-world database to assess attack success rates and mitigation effectiveness.

Main Results:

• Correlation attacks can significantly distort fingerprint bits (e.g., integrated attack distorts 64.8% of bits by modifying 14.2% of entries).
• Mitigation techniques drastically reduce the impact of correlation attacks, distorting only 3% of fingerprint bits.
• Mitigation remains effective even when attackers use accurate correlation models and higher-order correlations.

Conclusions:

• Existing database fingerprinting schemes are susceptible to correlation attacks.
• Developed mitigation techniques offer robust protection against various correlation attacks.
• The proposed methods are practical, enhancing security without compromising database utility.