Beyond the Math: Finding Flaws in Privacy-Preserving Systems
A new auditing framework reveals that even theoretically sound differential privacy implementations are vulnerable to subtle bugs that can compromise user data.
Science
Decoding Bose-Einstein Condensates in Quasiperiodic Landscapes
![Solutions [latex] u(x) [/latex] exhibit distinct patterns based on the number of nonzero symbols in their underlying codes; specifically, solutions with a single nonzero symbol differ structurally from those possessing two, a distinction visualized through their arrangement on a quasiperiodic lattice.](https://arxiv.org/html/2602.17172v1/x8.png)
Researchers have developed a novel coding approach to classify and understand the complex stationary states of Bose-Einstein condensates trapped in one-dimensional quasiperiodic lattices.
Atomic Views: Building Concurrent Data Structures with Read-Modify-Write
This research details how to create highly concurrent data structures using only basic read and write operations, providing a foundation for scalable shared memory systems.
Beyond English: Gauging AI Safety in South Asian Languages
A new benchmark reveals that current methods for evaluating AI vulnerability are often misleading when applied to languages beyond English, particularly in the diverse linguistic landscape of South Asia.
When Saying ‘No’ Isn’t Enough: Hardening Language Models Against Jailbreaks
New research reveals a critical flaw in current AI safety protocols – a tendency for models to override safety features when pressured – and proposes a more robust approach to prevent unintended outputs.
Hijacking AI Assistants: A New Era of Attack
Researchers have demonstrated a powerful new method for manipulating AI agents, exposing vulnerabilities beyond typical prompt injection defenses.
Fragile AI: Can Mamba Handle the Pressure in Medical Imaging?
![Vision Mamba proposes a selective state space model-a departure from traditional transformers-that utilizes a hardware-aware, temporally selective scan to process visual inputs, achieving a linear scaling in sequence length and promising improved efficiency in vision tasks through a principled reduction of computational redundancy inherent in attention mechanisms [latex]O(N) [/latex] versus [latex]O(N^2)[/latex].](https://arxiv.org/html/2602.16723v1/Fig/visionMamba.jpg)
A new study reveals that Mamba-based medical image analysis, despite its impressive performance, is surprisingly susceptible to both malicious attacks and real-world hardware failures.
Beyond Einstein: Mapping Multigravity with Black Hole Solutions
New research leverages the Kerr-Schild approach and the double copy to construct exact solutions in multigravity theories, revealing unexpected links to fundamental field interactions.
Beyond Accuracy: Measuring the true Strength of AI Reasoning
New research reveals that achieving high scores on reasoning tasks doesn’t guarantee a system’s ability to apply or justify its logic.
Black Hole Ripples: Unlocking the Secrets of Gravity’s Edge
New research reveals how the fundamental properties of black holes dictate their response to external forces, linking horizon behavior to observable tidal distortions.