Traduzido do inglês pela TechUpscale com ajuda de IA e verificado com o original. Ler o original em inglês
Novas pesquisas sobre IA e segurança, selecionadas pela relevância para modelos de linguagem, agentes de IA e cibersegurança. Os resumos são das próprias palavras dos autores ou da editora, com links para o original.
Revisado por pares
Layer-wise Prediction for Interpretation of Hidden Representations in Deep Neural Networks through Evolving Probability Distributions
Designing a blockchain-enabled certification framework for food loss and waste reduction: a proof-of-concept in the fresh food supply chain
Applying digital twins to prevent preterm births
Author Correction: Prediction of protein subcellular localization in single cells
Novos preprints (ainda não revisados por pares)
ICALL-BENCH: Benchmarking LLM Interprocedural Reasoning in Binary Programs
Recent benchmarks show that LLMs achieve strong performance in both function-level binary understanding and end-to-end binary reverse engineering tasks, but whether LLMs can reason over interprocedural semantics in compiled binaries remains unclear. Function-level benchmarks offer limited visibility into cross-function relationships, while whole-program benchmarks typically evaluate final outcomes of tasks that may be solvable from local clues alone, making it difficult to disentangle interprocedural reasoning from exploration, tool use, execution, and retrieval.
Jailbreak Connectivity: Towards Diverse, Transferable, and Universal Multimodal Large Language Model Jailbreak
Multimodal large language models (MLLMs) are vulnerable to visual jailbreak attacks that bypass safety safeguards. Existing methods typically optimize isolated adversarial examples, limiting attack diversity, transferability, and generalization.
SkillMem: Compact External Skill Memory for Latent Multi-Agent Reasoning
Latent multi-agent systems reduce inference cost by keeping intermediate reasoning and inter-agent communication in continuous representations. Reusable skills, however, are commonly supplied as text or stored as layer-wise Key-Value states or skill-specific model parameters, introducing repeated processing or substantial per-skill storage. We introduce textbf{SkillMem}, which represents each reusable skill as a small external latent memory and makes it directly available during latent multi-agent reasoning. A role-conditioned cross-attention reader allows the same memory to guide different agents according to their roles and current hidden states. We train the skill memories and shared reader through behavioral distillation, matching the change in token predictions induced by a textual skill while keeping the backbone frozen. For our experiments, we obtain textual skills from paired successful and failed multi-agent trajectories, although existing skills can also be compiled directly. Across six benchmarks and Qwen3 models ranging from 4B to 14B parameters, SkillMem provides strong gains for smaller backbones while preserving efficient latent multi-agent inference across model scales. We use the same $8times256$ memory configuration across all evaluated model scales, requiring 9.6,KB of skill-specific storage per skill, including its retrieval embedding. For a library of 100 skills, SkillMem reduces total storage by more than $38times$ compared with alternative skill representations and achieves up to a $4.23times$ end-to-end inference speedup. These results show that compact procedural memories can be reused without sacrificing the efficiency of latent multi-agent reasoning.
Evaluating Causal Reasoning in LLMs: A Case Study in the Challenges of Procedural Generation
Large language models (LLMs) appear to represent and reason about a large range of phenomena. In particular, recent studies have examined whether LLMs can accurately reason about causation.
WAAA! Web Adversaries Against Web Agents
Large language models (LLMs) are increasingly being integrated into web browsers to create agentic browsing systems that execute actions on behalf of the user. Prior work considering the security of agentic browsers focuses on indirect prompt-injection attacks.
Judging Before Propagation: Task-Graph Contextual Integrity for Multi-Agent Privacy
Multi-agent LLM systems increasingly rely on persistent memory and inter-agent communication. These capabilities create privacy risks that extend beyond individual interactions: a release appropriate in isolation may become inappropriate when combined with information previously disclosed to a recipient or propagated to downstream agents along the planned workflow.
PrivCode: When Code Generation Meets Differential Privacy
Large language models (LLMs) have presented outstanding performance in code generation and completion. However, fine-tuning these models on private datasets can raise privacy and proprietary concerns, such as the leakage of sensitive personal information.
Is it my preference? Role Grounding Attack against Agent's Memory via Social Interaction
LLM-based agents now share experiences on social platforms and remember new preferences to enhance their performance. However, existing memory systems contain a vulnerability in role-to-principal attribution that attackers can exploit.