AI Reinvents Fault Attacks: Deep Learning Takes on Lightweight Ciphers

In the ever-growing landscape of the Internet of Things (IoT), security remains a pressing concern. Lightweight cryptographic primitives, essential for these resource-constrained devices, face new threats from physical attacks, particularly the elusive fault attacks. The paper, published in Japanese, reveals a novel application of deep learning techniques to address these challenges.

Deep Learning Meets Fault Attacks

Recent advancements have shown deep learning can bolster cryptanalysis. However, its application to fault attacks, especially on stream ciphers, is still emerging. This research investigates the potential of deep learning-assisted differential fault attacks on three lightweight stream ciphers: ACORNv3, MORUSv2, and ATOM.

Using a relaxed fault model, researchers injected single-bit faults at unknown locations. They trained multilayer perceptron (MLP) models to pinpoint fault locations, achieving striking identification accuracies: 0.999880 for ACORNv3, 0.999231 for MORUSv2, and 0.823568 for ATOM. Compare these numbers side by side with traditional methods, and the advantage is clear.

Redefining Secret Recovery

The study doesn’t stop at identification. A threshold-based strategy was introduced to minimize fault injections needed for secret recovery. For ACORN, secrets were recoverable with just 21 to 34 faults. MORUS required more, between 213 to 248 faults, and up to 6 bits of guessing. Even so, these numbers substantially reduce attack complexity compared to previous approaches. But what about ATOM?

ATOM appears to boast a higher security margin. Most state bits in its Non-linear Feedback Shift Register (NFSR) can only be deciphered under a precise model. This highlights a key point: while deep learning shows promise, not all ciphers are equally vulnerable. Could this resilience make ATOM a new standard in secure IoT communications?

The Road Ahead

Crucially, these findings underscore the dual nature of AI in cryptography. While it enhances attack techniques, it equally drives the need for reliable defenses. The benchmark results speak for themselves, but they also serve as a call to action for developers and cryptographers alike.

Western coverage has largely overlooked this emerging trend. Yet, as AI continues to expand its cryptanalytic capabilities, the industry must adapt swiftly. Is the world of IoT ready for this new wave of AI-enhanced cryptanalysis? The stakes are high, and the clock is ticking.