Semantic Communication’s Unexpected Resilience
![Semantic communication systems demonstrate greater robustness against adversarial attacks than classical systems, consistently requiring a higher minimum attack power [latex]\rho^{\*}[/latex] to achieve comparable distortion in image transmission-a testament to their inherent resilience.](https://arxiv.org/html/2603.24082v1/x5.png)
New research reveals that semantic communication systems demonstrate surprising robustness against adversarial attacks, challenging conventional wisdom about their vulnerability.
Science
Building Trust in AI Agent Networks
As AI agents proliferate, ensuring the security of their interactions is paramount, and this paper reveals hidden vulnerabilities when these protocols are combined.
Beyond the Ground: Securing Space Systems with Decentralized Trust
A new architecture is proposed to relocate the root of trust to orbital platforms, bolstering the security of satellite networks and the workloads they support.
Beyond the Limit: Building Robust Networks with Faster Distance Queries
Researchers have devised a new method for creating fault-tolerant distance oracles and spanners that dramatically reduces space complexity without sacrificing accuracy.
Beyond Pixels: Reinforcement Learning Powers Reliable Image Transmission
![The proposed JS3C framework establishes a foundation for systems to navigate inevitable decay, prioritizing sustained functionality over resisting the passage of time-a principle akin to accepting [latex] \Delta t [/latex] as an inherent property of existence rather than an adversary to be overcome.](https://arxiv.org/html/2603.23869v1/x1.png)
A new approach to semantic communication utilizes adaptive retransmission strategies to drastically improve the reliability and efficiency of image delivery over challenging network conditions.
Squeezing Secrets From Tiny AI: The Risks to Edge Language Models
![The study delineates a threat framework for knowledge extraction from large language models deployed on edge devices, highlighting that despite operating under quantization-such as [latex]INT4[/latex] or [latex]INT8[/latex]-and resource constraints that limit query budgets and introduce noise, strategically designed queries can still elicit substantial behavioral knowledge from these models, contrasting with traditional extraction methods reliant on full-precision teacher models and abundant computational resources.](https://arxiv.org/html/2603.23822v1/x2.png)
As powerful language models shrink to run on devices like phones and IoT sensors, a new study reveals that reducing their size doesn’t necessarily protect the valuable knowledge they contain.
Shaping Secrets: A New Path to Post-Quantum Encryption
Researchers are exploring how the geometry of complex 3D shapes can underpin more secure key exchange mechanisms in the age of quantum computing.
Securing 5G’s Wireless Edge with Quantum Encryption
Researchers demonstrate a practical approach to bolstering 5G network security for non-3GPP devices by integrating Quantum Key Distribution with standard IPsec protocols.
Beyond the Crystal: Unveiling Hidden Order in Quantum Magnets
![The implementation of the [latex]\pi\pi[/latex]-flux condition within the investigated gauge for the IGT relies on a two-site unit cell comprising sublattices designated AA and BB, where the sign of [latex]\rho^{z}\_{ij}[/latex]-indicated by dashed (red, [latex]\rho^{z}\_{ij}=-1[/latex]) or solid (black, [latex]\rho^{z}\_{ij}=+1[/latex]) links-determines the specific gauge choice.](https://arxiv.org/html/2603.23154v1/Figures/two-site-unit-cell.png)
New research demonstrates a pathway from seemingly simple magnetic models to exotic states of matter exhibiting both crystalline and topological order.
Hidden in Plain Sight: A New Approach to Private Data Aggregation
![A compressed two-layer privacy protocol demonstrates an inherent trade-off between data privacy and utility, revealing that minimizing the privacy parameter [latex]\varepsilon[/latex] to approximately 13-achieved with nine decoys-necessitates operating at a signal-to-noise ratio near one, where the signal is comparable to noise, and results in only a marginal 0.3% reduction in prior uncertainty as measured by normalized mean-squared error.](https://arxiv.org/html/2603.22808v1/x1.png)
Researchers have developed a novel framework, PolyVeil, that leverages geometric principles to securely combine private data streams without relying on conventional cryptographic methods.