Stress-Testing AI Agents: A New Approach to Security

Researchers have developed an adaptive architecture for systematically evaluating and improving the safety of complex AI agents against adversarial attacks.

Researchers have developed an adaptive architecture for systematically evaluating and improving the safety of complex AI agents against adversarial attacks.
![The study precisely determines [latex]\delta(0)[/latex] within Quantum Chromodynamics (QCD) through higher-order coefficient analysis, demonstrating that perturbative expansions-with uncertainties represented by shaded regions-converge towards a mean value-defined by the Shanks sum-while variations across different resummation techniques, visualized as yellow bands, quantify the inherent spread around this central prediction, all calculated using [latex]\alpha_s = 0.31959[/latex].](https://arxiv.org/html/2601.11277v1/x1.png)
Researchers are leveraging advanced mathematical techniques to refine calculations of hadronic tau decays, a crucial process for pinpointing the strength of the strong force.
![Yb atoms exhibit Kondo physics through a carefully engineered system where fermionic atoms in spin-1/2 states are selectively trapped - mobile in a ground band and localized in an excited band - creating an impurity interacting with a surrounding fermionic environment via both antiferromagnetic and potential interactions, effectively modeling [latex]Eq. (1)[/latex] and demonstrating how these interactions draw atoms either towards or away from the impurity site, as visualized by the wavefunction modulus.](https://arxiv.org/html/2601.11449v1/x1.png)
New research reveals that the Kondo effect, a hallmark of quantum impurity problems, can persist in one-dimensional systems even with strong scattering, thanks to the crucial role of environmental correlations.
New research demonstrates how zero-knowledge proofs can establish trust in hardware components without revealing sensitive design details.

A new framework embeds security directly within the interposer layer, enabling distributed authentication for complex, multi-chip packages.

A new study rigorously tests the accuracy of modern density functional theory methods in predicting the structural behavior of group IV elements like silicon and germanium.
![Performance metrics demonstrate the capabilities of [latex]\mathsf{POD}[/latex] when operating with the eBCH(16,7)(16,7) coding scheme, indicating its efficacy within that specific communication architecture.](https://arxiv.org/html/2601.11373v1/fig/eBCH_4_2_POD.png)
A new decoding framework leverages the underlying symmetries of polar codes to significantly improve speed and performance in error correction.
![The proposed LoRA oracle efficiently adapts large language models by learning low-rank approximations, enabling parameter-efficient transfer learning without extensive retraining of the original model weights [latex] \Delta w = BA [/latex], where [latex] B [/latex] and [latex] A [/latex] represent the low-rank matrices and [latex] w [/latex] denotes the original weights.](https://arxiv.org/html/2601.11207v1/x1.png)
A new framework efficiently probes deep learning models to reveal training data origins and hidden security vulnerabilities.

A novel quantum neural network architecture demonstrates improved anomaly detection in smart grid systems, offering enhanced resilience against increasingly sophisticated cyber threats.
New research reveals a fundamental link between quantum communication and the mathematical structures governing semantic alignment, demonstrating how entanglement can minimize the cost of shared understanding.