redeval

RedEval - LLM 안전성 평가 프레임워크 체계적인 공격attack 및 거부refusal 테스트를 통해 대규모 언어 모델LLM의 안전성을 평가하기 위한 포괄적인 프레임워크입니다. RedEval은 적대적 프롬프트와 유해 콘텐츠에 대한 LLM의 견고성을 평가하기 위한 통합되고 안전하며 확장 가능한 플랫폼을 제공합니다. 이는 LLM의 종합적인 레드팀링red teaming을 위한 범용 데이터셋인 논문 "RedBench: A Universal Dataset for Comprehensive Red Teaming of Large...

gemini-2.5-pro-nf-tables-red-teaming

Gemini 2.5 Pro nftables レッドチーミング事例研究 CVE-2023-32233 LLM安全性研究 | 責任ある開示 | AIアライメント評価 研究者 : Niranj R Mahaswar Destawell Google AI脆弱性報奨金プログラム : 889286 — 対象外 概要 このリポジトリは、Linuxカーネルの脆弱性 CVE-2023-32233 （nftables レースコンディション / Use-After-Free）に関する技術的プリミティブについて、フロンティアモデルを対象に行った比較 LLMレッドチーミング 実験を記録したものです。...

gemini-2.5-pro-nf-tables-red-teamin

gemini-2.5-pro-nf-tables-red-teamin Uno studio di caso tecnico e un dataset temporale che documentano le politiche di allineamento alla sicurezza, le protezioni e l'evoluzione del comportamento di rifiuto di Google Gemini 2.5 Pro riguardo alle primitive legacy di vulnerabilità del kernel Linux...

Astra Linux – Vulnerability found in Linux 5.15, Linux 6.1

In the Linux kernel, the following vulnerability has been resolved: remoteproc: mediatek: Ensure that the IPI buffer fits within the L2TCM. The location of the IPI buffer is determined from the firmware that we load into the System Companion Processor. It’s not guaranteed that both the SRAM size...

A New Framework for Cybersecurity Refusals in AI Agents

Agentic scaffolds have dramatically improved LLM performance on complex, long-horizon tasks, yielding both broad benefits and amplified risks in domains like cybersecurity. Existing benchmarks for AI agents in cybersecurity focus mainly on measuring proficiency--how effectively agents can complet...

Refusal Before Decoding: Detecting and Exploiting Refusal Signals in Intermediate LLM Activations

In this paper, we investigate whether refusal behavior can be predicted from LLM intermediate activations before decoding using linear probes trained on residual stream activations at each transformer block. We find that refusal is linearly decodable well before the final layer, indicating that...

Linux kernel 安全漏洞

The Linux kernel is the core of the open-source operating system Linux, developed by the Linux Foundation in the United States. There is a security vulnerability in the Linux kernel, which stems from the smcclcwaitmsg function accessing the link group state prematurely when a CLC refusal occurs...

Refusal Evaluation in Coding LLMs and Code Agents: A Systematic Review of Thirteen Malicious-Code Prompt Corpora (2023-2025)

The evaluation of large language model refusal on malicious-coding tasks now spans at least thirteen publicly released prompt corpora AdvBench, the CyberSecEval family, RMCBench, RedCode, MCGMark, JailbreakBench, CySecBench, MalwareBench, CIRCLE, MOCHA, ASTRA, Scam2Prompt / Innoc2Scam-bench, and...

Ablating Safety: Mechanisms for Removing Alignment in Language Models for Security Applications

Safety-aligned language models often refuse cybersecurity requests whose wording resembles misuse, even when the task is authorized and defensive. This makes security evaluation ambiguous: a failed answer may reflect missing capability or refusal-policy intervention. Ablating Safety studies...

Tracing the Dynamics of Refusal: Exploiting Latent Refusal Trajectories for Robust Jailbreak Detection

Representation Engineering typically relies on static refusal vectors derived from terminal representations. We move beyond this paradigm, demonstrating that refusal is a dynamic and sparse process rather than a localized outcome. Using Causal Tracing, we uncover the Refusal Trajectory-a persiste...

Not All Tokens Are Created Equal: Query-Efficient Jailbreak Fuzzing for LLMs

Large Language ModelsLLMs are widely deployed, yet are vulnerable to jailbreak prompts that elicit policy-violating outputs. Although prior studies have uncovered these risks, they typically treat all tokens as equally important during prompt mutation, overlooking the varying contributions of...

Activation Surgery: Jailbreaking White-Box LLMs without Touching the Prompt

Most jailbreak techniques for Large Language Models LLMs primarily rely on prompt modifications, including paraphrasing, obfuscation, or conversational strategies. Meanwhile, abliteration techniques also known as targeted ablations of internal components have been used to study and explain LLM...

RedBench: A Universal Dataset for Comprehensive Red Teaming of Large Language Models

As large language models LLMs become integral to safety-critical applications, ensuring their robustness against adversarial prompts is paramount. However, existing red teaming datasets suffer from inconsistent risk categorizations, limited domain coverage, and outdated evaluations, hindering...

About the security content of Compressor 4.11.1

About the security content of Compressor 4.11.1 This document describes the security content of Compressor 4.11.1. About Apple security updates For our customers' protection, Apple doesn't disclose, discuss, or confirm security issues until an investigation has occurred and patches or releases ar...

SUSE CVE-2023-53695

In the Linux kernel, the following vulnerability has been resolved: udf: Detect system inodes linked into directory hierarchy When UDF filesystem is corrupted, hidden system inodes can be linked into directory hierarchy which is an avenue for further serious corruption of the filesystem and kerne...

UBUNTU-CVE-2023-53695

In the Linux kernel, the following vulnerability has been resolved: udf: Detect system inodes linked into directory hierarchy When UDF filesystem is corrupted, hidden system inodes can be linked into directory hierarchy which is an avenue for further serious corruption of the filesystem and kerne...

EUVD-2008-5375

Malware in sbrugna...

EUVD-2017-17963

Malware in sbrugna...

EUVD-2024-54426

Malicious code in bioql PyPI...

Beyond Surface Alignment: Rebuilding LLMs Safety Mechanism Via Probabilistically Ablating Refusal Direction

Jailbreak attacks pose persistent threats to large language models LLMs. Current safety alignment methods have attempted to address these issues, but they experience two significant limitations: insufficient safety alignment depth and unrobust internal defense mechanisms. These limitations make...