Fragile Guardians: How Easily Code Attacks Slip Past AI Security
New research reveals that current AI-powered vulnerability detectors are surprisingly susceptible to evasion through subtle code transformations that preserve functionality.
Science
Keeping Distributed Systems Honest: A Resilient Termination Detector
This research introduces a fault-tolerant adaptation of Safra’s algorithm, guaranteeing reliable detection of termination even when nodes fail.
Predicting Nuclear Decay with a New Computational Approach
![The distribution of deviations between calculated and experimental α-decay half-lives, assessed across the chart of nuclides, reveals the sensitivity of predictive models-specifically those employing the WS potential with depth [latex]V_0[/latex] determined via both the BSQC method and fitted parametrization-to accurately forecasting nuclear stability, a sensitivity further illuminated by comparison to semi-microscopic calculations utilizing the CDM3Y3 interaction, and subtly demarcated by the influence of magic numbers [latex]N=126[/latex] and [latex]Z=82[/latex].](https://arxiv.org/html/2602.02070v1/x2.png)
Researchers have developed a faster and more efficient method for calculating the half-lives of alpha-decaying nuclei, offering a valuable tool for nuclear physics and astrophysics.
Decoding Quantum Noise: A New Entropy Framework for Secure Communication
Researchers have developed a unified approach to understanding and quantifying noise in quantum communication channels, paving the way for more robust and secure data transmission.
Stealing Secrets While Your System Sleeps
Researchers have discovered a novel power side-channel attack that exploits operating system behavior to compromise the security of cryptographic keys.
Squeezing More from Less: A New Approach to Model Quantization
![Structured Residual Reconstruction mitigates the damage caused by weight quantization by proactively preserving dominant low-rank structures, resulting in a significantly reduced reconstruction error-measured as [latex]\lVert\mathbf{W}-\mathbf{Q}-\mathbf{L}\mathbf{R}\rVert\_{F}[/latex]-compared to standard quantization error reduction techniques that destroy this intrinsic structure.](https://arxiv.org/html/2602.02001v1/x1.png)
Researchers have developed a method for dramatically reducing the size of large language models without sacrificing performance by strategically preserving key information and reconstructing lost details.
Securing the 6G Road: Adaptive Quantum-Safe Crypto for Connected Cars
As vehicles become increasingly connected, this review explores how to dynamically deploy post-quantum cryptography to safeguard V2X communications against future threats.
Taming Spectre: A Formally Verified Defense
Researchers have developed a new security approach, SpecIBT, to neutralize Spectre attacks by combining hardware-assisted control-flow integrity with compiler optimizations.
Symmetry’s Shield: Protecting Order in Quantum Systems
New research demonstrates how higher-order symmetries can stabilize quantum systems against disorder, effectively raising the bar for spontaneous symmetry breaking.
Squeezing More From Less: A New Approach to Efficient AI Models
![LoRDS reimagines parameter-efficient fine-tuning by decomposing block-wise scaling into a low-rank product [latex]W\odot B_{AW} [/latex], enabling granular refinement via post-training quantization or direct fine-tuning with high-rank updates and-unlike methods like QLoRA which introduce additive, non-mergeable adapters-zero additional inference overhead through natural absorption into the dequantization process.](https://arxiv.org/html/2601.22716v1/x2.png)
Researchers have developed a novel method for compressing and adapting large language models, boosting performance and reducing computational demands.