Tootfinder

MOCHA: Are Code Language Models Robust Against Multi-Turn Malicious Coding Prompts?
Muntasir Wahed, Xiaona Zhou, Kiet A. Nguyen, Tianjiao Yu, Nirav Diwan, Gang Wang, Dilek Hakkani-T\"ur, Ismini Lourentzou
https://arxiv.org/abs/2507.19598

MOCHA: Are Code Language Models Robust Against Multi-Turn Malicious Coding Prompts?
Recent advancements in Large Language Models (LLMs) have significantly enhanced their code generation capabilities. However, their robustness against adversarial misuse, particularly through multi-turn malicious coding prompts, remains underexplored. In this work, we introduce code decomposition attacks, where a malicious coding task is broken down into a series of seemingly benign subtasks across multiple conversational turns to evade safety filters. To facilitate systematic evaluation, we int…

SDD: Self-Degraded Defense against Malicious Fine-tuning
Zixuan Chen, Weikai Lu, Xin Lin, Ziqian Zeng
https://arxiv.org/abs/2507.21182 https://arxiv.org/pd…

SDD: Self-Degraded Defense against Malicious Fine-tuning
Open-source Large Language Models (LLMs) often employ safety alignment methods to resist harmful instructions. However, recent research shows that maliciously fine-tuning these LLMs on harmful data can easily bypass these safeguards. To counter this, we theoretically uncover why malicious fine-tuning succeeds and identify potential defense strategies. Building on the theoretical analysis, we introduce the Self-Degraded Defense (SDD) framework. SDD encourages LLMs to produce high-quality but irr…

"We simply don’t know to defend against these attacks. We have zero agentic AI systems that are secure against these attacks.[…] It’s an existential problem that, near as I can tell, most people developing these technologies are just pretending isn’t there."
https://www.

Este es un buen artículo de opinión para entender cómo la administración de Rodrigo Chaves Robles (alias zoodrigo, el presidente de Costa Rica acosador sexual que llegó al poder violando leyes de financiamiento de campañas políticas y parece gobernar para no estorbar al narco) ha hecho un daño enorme a instituciones públicas "desde adentro": poniendo a jerarcas "sumisos y servirles" en puestos de juntas directivas de instituciones autónomas para que tomen decisiones contr…

Very high-energy gamma-ray and neutrino emission from hadronic interaction in compact binary millisecond pulsars
Vittoria Vecchiotti, Manuel Linares
https://arxiv.org/abs/2508.20952

Very high-energy gamma-ray and neutrino emission from hadronic interaction in compact binary millisecond pulsars
Blackwidow and redback systems are millisecond pulsars in compact orbits with ultra-light and low-mass companions, respectively, collectively known as ``spider pulsars". In such systems, an intrabinary shock can form between the pulsar and the companion winds, serving as a site for particle acceleration and associated non-thermal emission. Assuming that protons can be extracted from the neutron star surface and accelerated at the intrabinary shock and/or within the pulsar wind, we model the ver…

Mir wird DIE ZEIT ja schon seit längerem immer etwas fremder. Aber gut, man entwickelt sich und vielleicht verkläre ich die Zeitung "von früher" auch nur nostalgisch. Bisher war ich froh, dass es sie als Stimme gibt.
Aber bei einem der wichtigsten Themen überhaupt - der Zukunft unserer Demokratie - wirkt allein schon die Bildsprache inzwischen so plump und ungelenk wie vom TikTok Clickbait-Slop abgeschaut. Die Zeitung und der Diskurs brauchen solche Dämlichkeiten nicht.

📢 Waffenruhe zwischen Thailand und Kambodscha vereinbart
Thailand und Kambodscha haben sich auf eine sofortige Feuerpause geeinigt. Das teilte der malaysische Ministerpräsident Anwar Ibrahim nach Verhandlungen der beiden Länder nach den jüngsten Auseinandersetzungen mit.
➡️ https://www.

Waffenruhe zwischen Thailand und Kambodscha vereinbart
Thailand und Kambodscha haben sich auf eine sofortige Feuerpause geeinigt. Das teilte der malaysische Ministerpräsident Anwar Ibrahim nach Verhandlungen der beiden Länder nach den jüngsten Auseinandersetzungen mit.

Guarding Against Malicious Biased Threats (GAMBiT) Experiments: Revealing Cognitive Bias in Human-Subjects Red-Team Cyber Range Operations
Brandon Beltz, Jim Doty, Yvonne Fonken, Nikolos Gurney, Brett Israelsen, Nathan Lau, Stacy Marsella, Rachelle Thomas, Stoney Trent, Peggy Wu, Ya-Ting Yang, Quanyan Zhu
https://arxiv.org/abs/2508.20963…

Guarding Against Malicious Biased Threats (GAMBiT) Experiments: Revealing Cognitive Bias in Human-Subjects Red-Team Cyber Range Operations
We present three large-scale human-subjects red-team cyber range datasets from the Guarding Against Malicious Biased Threats (GAMBiT) project. Across Experiments 1-3 (July 2024-March 2025), 19-20 skilled attackers per experiment conducted two 8-hour days of self-paced operations in a simulated enterprise network (SimSpace Cyber Force Platform) while we captured multi-modal data: self-reports (background, demographics, psychometrics), operational notes, terminal histories, keylogs, network packe…

An energy approach to pulsar-disc interaction: disc stability and implications for transitional millisecond pulsars
Eda Vurgun, Domingo Garc\'ia-Senz, Manuel Linares, K. Yavuz Eksi
https://arxiv.org/abs/2505.23407

An energy approach to pulsar-disc interaction: disc stability and implications for transitional millisecond pulsars
The stability of an accretion disc surrounding a millisecond pulsar is analysed from an energetic point of view, using magnetohydrodynamic simulations that consider realistic disc structures and a variety of magnetic field inclination angles. The time-averaged components of the magnetic field interact with the disc through ohmic dissipation, which causes heating and partial evaporation of its innermost region. The stability of the disc right after the magnetic field is turned on is analysed as …

AgentAlign: Navigating Safety Alignment in the Shift from Informative to Agentic Large Language Models
Jinchuan Zhang, Lu Yin, Yan Zhou, Songlin Hu
https://arxiv.org/abs/2505.23020

AgentAlign: Navigating Safety Alignment in the Shift from Informative to Agentic Large Language Models
The acquisition of agentic capabilities has transformed LLMs from "knowledge providers" to "action executors", a trend that while expanding LLMs' capability boundaries, significantly increases their susceptibility to malicious use. Previous work has shown that current LLM-based agents execute numerous malicious tasks even without being attacked, indicating a deficiency in agentic use safety alignment during the post-training phase. To address this gap, we propose AgentAlign, a novel framework t…