Hijacking the Hand-Off: Securing AI Routing Systems
As large language models become integrated into complex workflows, the systems that direct those models are increasingly vulnerable to attack.
Science
Beyond the Toric Code: Mapping Self-Dual Transitions
![The study of a deformed toric code reveals a “multi-critical” point-where two Ising transitions converge on the self-dual line [latex]h_x = h_z[/latex]-leading to a topologically trivial phase characterized by spontaneous self-duality symmetry breaking, demonstrating how complex systems can transition between ordered states through precisely defined critical configurations.](https://arxiv.org/html/2601.20945v1/x1.png)
A new theoretical framework reveals a deeper connection between self-dual Higgs transitions and the emergence of exotic topological orders.
Charmed Decay Hints at Standard Model Consistency
![The analysis simultaneously fits measured decay rates and forward-backward asymmetries of [latex]D^{0(+)}\to\bar{K}\ell^{+}\nu_{\ell}[/latex], demonstrating that inclusion of scalar current contributions is crucial for accurately modeling the ratios of differential decay rates [latex]\mathcal{R}_{\mu/e}[/latex].](https://arxiv.org/html/2601.21185v1/x754.png)
New measurements of D meson decays are providing crucial constraints on fundamental parameters and testing the limits of lepton flavor universality.
Learning to Keep Systems Safe Without Knowing How They Work
![A safety filter, learned solely from system transitions, employs a safety critic to map states and constraints to safety valuations and their derivatives, which then inform a quadratic programming solver-in conjunction with initial reference inputs-to generate demonstrably safe control actions, as evidenced by its ability to stabilize an inherently unstable system even when subjected to aggressive [latex] \pm 1 [/latex] reference commands.](https://arxiv.org/html/2601.21297v1/Figures/Thumbnail_v5.png)
A new data-driven approach enables safe control of complex systems, even when their internal dynamics are unknown.
Beyond Finite Fields: Scaling Symbolic Security Verification
Researchers have expanded the capabilities of automated protocol analysis to encompass the full algebraic power of Diffie-Hellman groups, unlocking more robust security proofs.
Shared Secrets: Securing Digital Signatures for the Post-Quantum Era
A new approach distributes the risk and responsibility of cryptographic key management, bolstering security for next-generation digital signatures.
Beyond Local Minima: Certifying Solutions for Robust PCA
![Robust principal component analysis successfully recovers underlying low-rank plus sparse structure in synthetic matrices of size 100x80 with high probability, achieving a recovery success rate consistently exceeding 99.9% as quantified by a normalized reconstruction error of less than [latex]10^{-3}[/latex] across 10,000 independent trials.](https://arxiv.org/html/2601.21333v1/x4.png)
New research establishes rigorous guarantees for finding optimal solutions in robust principal component analysis, even when dealing with challenging, non-convex problems.
Beyond Privacy: Designing Neural Networks for Encrypted Computation
A new architecture minimizes the performance overhead of homomorphic encryption, paving the way for faster and more practical privacy-preserving machine learning.
Securing the Research Pipeline: A New Framework for Software Supply Chain Security
A standardized taxonomy for research software supply chains is crucial for consistently evaluating vulnerabilities and mitigating risks in academic and scientific computing.
Checking Its Work: A New Approach to Truthful AI
Researchers have developed a decoding framework that empowers large language models to self-assess and refine their outputs, dramatically reducing factual errors.