Affichage de 17 article(s)
[High] LogJack demonstrates indirect prompt injection in LLM debugging agents via cloud log content, enabling command execution and remote code execution. (Source: ArXiv cs.CR, https://arxiv.org/ab...
[Notable] The Synthetic Media Shift introduces the CONVEX dataset of multimodal misinformation and shows AI-generated content exhibits disproportionate virality and detectability challenges. (Sourc...
[Disruptive] HarmfulSkillBench benchmarks agent safety against harmful skills across two registries, showing that 4.93% of analyzed skills could be weaponized. (Source: ArXiv cs.AI, https://arxiv.o...
[Notable] Grinex, a US-sanctioned currency exchange, claims that a $15 million heist was conducted by unfriendly states and that such states have access to hack capabilities. (Source: Ars Technica ...
[Low] The article presents a locally deployable, agentic workflow that combines a rule-based extractor (Presidio) for structured PII and a domain-adapted, f… (Source: ArXiv cs.CR, https://arxiv.org...
[Notable] LinuxArena provides a large, diverse control setting for AI agents operating in live production environments, enabling sabotage and monitoring evaluat…
[High] A Systematization of Knowledge identifying 12 cross-layer attack vectors in autonomous LLM agents used for commerce and finance, and proposing mitigations. (Source: ArXiv cs.CR, https://arxi...
[Critical] Successful exploitation could allow an unauthenticated attacker to modify simulation parameters, training configuration and training data. (Source: CISA Advisories, https://www.cisa.gov/...
[Medium] Microsoft is rolling out File Explorer changes in Windows 11 Insider builds to improve launch speed and overall performance. (Source: BleepingComputer, https://www.bleepingcomputer.com/new...
[Disruptive] The article highlights which Big Tech players are winning the race to transition to post-quantum cryptography. (Source: Ars Technica AI, https://arstechnica.com/security/2026/04/while-...
The BAIR Blog introduces SPEX and ProxySPEX, ablation-based methods to identify influential interactions in large language models by perturbing inputs.
A BAIR Blog post describes a framework for evaluating and optimizing imaging systems based on mutual information, enabling direct assessment of information flows.
Successful exploitation of this vulnerability could allow an attacker to gain unauthorized access to systems and services.
Microsoft warned that a recent Microsoft Edge browser update introduced a bug that breaks right-click paste in chats in the Microsoft Teams desktop client.
As Section 702 surveillance law nears expiration, lawmakers remain divided over the 2024 changes and what they mean for U.S. surveillance authorities.
Copyright © DevOps Shield. Tous droits réservés. Politique de confidentialité | Politique de témoins | Conditions d'utilisation