In brief
- Introduces novel approaches in side-channel attacks by manipulating power signals or GND radiation.
- Explores privacy protection for language models and network anomaly detection using advanced private set intersection protocols.
- Demonstrates practical and automated post-quantum end-to-end tunnels without per-server client configuration.
Power-Related Side-Channel Attacks using the Android Sensor Framework
Mobile devices expose numerous power-related signals beyond timing proxies that allow for power side-channel attacks through unprivileged sensors provided by the Android sensor framework.
Identifies a novel leakage primitive, the rotation-dependent power leakage of the geomagnetic rotation vector sensor, achieving correlations greater than 0.9 for some tested sensors.
Evaluates the capabilities of the identified leakage primitives via two case studies: remote attacker via Chrome browser, and local attacker running inside an installed app, enabling an end-to-end pixel-stealing attack.
Mathias Oberhuber, Martin Unterguggenberger, Lukas Maar, Andreas Kogler, Stefan Mangard. “Power-Related Side-Channel Attacks using the Android Sensor Framework.” NDSS — https://www.ndss-symposium.org/ndss-paper/power-related-side-channel-attacks-using-the-android-sensor-framework/
PowerRadio: Manipulate Sensor Measurement via Power GND Radiation
Introduces PowerRadio, a new threat vector that bypasses existing protections by leveraging interconnected ground (GND) wires to inject malicious signals, changing sensor readings from a distance.
The injected signal couples with the sensor’s analog measurement wire and survives noise filters, manipulating sensors by inducing static bias, periodical signals, or pulses.
Root causes are identified as the lack of shielding between ground and data signal wires, and the asymmetry of circuit impedance enabling interference to bypass filtering.
Yan Jiang, Xiaoyu Ji, Yancheng Jiang, Kai Wang, Chenren Xu, Wenyuan Xu. “PowerRadio: Manipulate Sensor Measurement via Power GND Radiation.” NDSS — https://www.ndss-symposium.org/ndss-paper/powerradio-manipulate-sensor-measurement-via-power-gnd-radiation/
PQConnect: Automated Post-Quantum End-to-End Tunnels
Introduces PQConnect, a post-quantum end-to-end tunneling protocol that automatically protects packets between installed clients and servers without requiring changes to higher-level protocols.
Automatically creates end-to-end tunnels using automatic peer discovery via server announcements, requiring no client-side per-server configuration.
Uses a conservative post-quantum KEM for the server’s long-term public key to encrypt and authenticate all packets, and a smaller post-quantum KEM for forward secrecy alongside elliptic curves.
Daniel J. Bernstein, Tanja Lange, Jonathan Levin, Bo-Yin Yang. “PQConnect: Automated Post-Quantum End-to-End Tunnels.” NDSS — https://www.ndss-symposium.org/ndss-paper/pqconnect-automated-post-quantum-end-to-end-tunnels/
Prεεmpt: Sanitizing Sensitive Prompts for LLMs
Addresses the problem of formally protecting sensitive information contained in an LLM prompt while maintaining response quality by introducing a prompt sanitizer concept.
Proposes Prεεmpt, a system that categorizes sensitive tokens: format-dependent tokens (e.g. SSNs) protected via format-preserving encryption, and value-dependent tokens (e.g. age) protected via metric differential privacy.
Demonstrates that Prεεmpt achieves meaningful privacy guarantees while maintaining high utility compared to unsanitized prompts.
Amrita Roy Chowdhury, David Glukhov, Divyam Anshumaan, Prasad Chalasani, Nicholas Papernot, Somesh Jha, Mihir Bellare. “Prεεmpt: Sanitizing Sensitive Prompts for LLMs.” NDSS — https://www.ndss-symposium.org/ndss-paper/pr%ce%b5%ce%b5mpt-sanitizing-sensitive-prompts-for-llms/
Practical Traceable Over-Threshold Multi-Party Private Set Intersection
Proposes Practical Traceable Over-Threshold Multi-Party Private Set Intersection (MP-PSI) to compute the intersection over elements held by at least a predefined number of participants while supporting traceability.
Existing MP-PSI with threshold protocols provide full anonymity, which is problematic for scenarios like network anomaly detection and digital forensics where establishing provenance and localizing alerts is essential.
Le Yang, Weijing You, Huiyang He, Kailiang Ji, Jingqiang Lin. “Practical Traceable Over-Threshold Multi-Party Private Set Intersection.” NDSS — https://www.ndss-symposium.org/ndss-paper/practical-traceable-over-threshold-multi-party-private-set-intersection/
PriSrv+: Privacy and Usability-Enhanced Wireless Service Discovery with Fast and Expressive Matchmaking Encryption
Introduces PriSrv+, an advanced privacy and usability-enhanced service discovery protocol addressing limitations in expressiveness, privacy, scalability, and efficiency of PriSrv (NDSS'24).
Develops Fast and Expressive Matchmaking Encryption (FEME), a matchmaking encryption scheme capable of supporting expressive access control policies with an unbounded attribute universe.
Optimizes cryptographic operations, achieving 7.62x faster encryption and 6.23x faster decryption compared to PriSrv, while reducing ciphertext sizes by 87.33%.
Yang Yang, Guomin Yang, Yingjiu Li, Pengfei Wu, Rui Shi, Minming Huang, Jian Weng, HweeHwa Pang, Robert H. Deng. “PriSrv+: Privacy and Usability-Enhanced Wireless Service Discovery with Fast and Expressive Matchmaking Encryption.” NDSS — https://www.ndss-symposium.org/ndss-paper/prisrv-privacy-and-usability-enhanced-wireless-service-discovery-with-fast-and-expressive-matchmaking-encryption/
Privacy-Preserving Data Deduplication for Enhancing Federated Learning of Language Models
Introduces Efficient Privacy-Preserving Multi-Party Deduplication (EP-MPD), a protocol that efficiently removes duplicates from multiple clients’ datasets without compromising data privacy.
Constructed in a modular fashion utilizing two novel variants of the Private Set Intersection protocol to address scalability and potential privacy violations in federated setups.
Experiments demonstrate up to 19.62% improvement in perplexity and up to 27.95% reduction in running time while varying the duplication level between 10% and 30%.
Aydin Abadi, Vishnu Asutosh Dasu, Sumanta Sarkar. “Privacy-Preserving Data Deduplication for Enhancing Federated Learning of Language Models.” NDSS — https://www.ndss-symposium.org/ndss-paper/privacy-preserving-data-deduplication-for-enhancing-federated-learning-of-language-models/
PrivATE: Differentially Private Average Treatment Effect Estimation for Observational Data
Introduces PrivATE, a practical average treatment effect (ATE) estimation framework for observational data that ensures differential privacy.
Accommodates different privacy requirements by designing two levels of privacy protection (label-level and sample-level).
Derives an adaptive matching limit to balance noise-induced error and matching error, leading to a more accurate estimate of ATE.
Quan Yuan, Xiaochen Li, Linkang Du, Min Chen, Mingyang Sun, Yunjun Gao, Shibo He, Jiming Chen, Zhikun Zhang. “PrivATE: Differentially Private Average Treatment Effect Estimation for Observational Data.” NDSS — https://www.ndss-symposium.org/ndss-paper/private-differentially-private-average-treatment-effect-estimation-for-observational-data/