94 matches found
redeval
RedEval - LLM Safety Evaluation Framework A comprehensive framework for evaluating the safety of Large Language Models LLMs through systematic attack and refusal testing. RedEval provides a unified, secure, and extensible platform for assessing LLM robustness against adversarial prompts and harmf...
gemini-2.5-pro-nf-tables-red-teaming
Gemini 2.5 Pro nftables Red Teaming Case Study CVE-2023-32233 LLM Safety Research | Responsible Disclosure | AI Alignment Evaluation Researcher : Niranj R Mahaswar Destawell Google AI Vulnerability Reward Program : 889286 — Out of Scope Summary This repository documents a comparative LLM red...
CVE-2026-93710
Dancer2 versions from 2.0.0 before 2.2.0 for Perl dispatch a route that a dying hook refused when the exception handler halts the response in compilehooks. A hook that dies fires core.app.hookexception, then calls cleanup unless the failing hook is the exception handler. A handler that halts does...
gemini-2.5-pro-nf-tables-red-teamin
gemini-2.5-pro-nf-tables-red-teamin A technical case study and timeline dataset documenting Google Gemini 2.5 Pro's safety alignment policies, guardrails, and refusal behavior evolution regarding legacy Linux kernel vulnerability primitives CVE-2023-32233. Link to case study website :...
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...
OpenAI Launches GPT-5.6-Cyber with Reduced Safeguards for Exploit Development
OpenAI on Monday unveiled a new cybersecurity-focused model called GPT‑5.6‑Cyber that it said is focused on vulnerability research, penetration testing, and incident response. "Built on GPT‑5.6 Sol, it is trained to improve capabilities on several specialized cybersecurity tasks e.g., finding...
📄 WordPress Fluent Forms 6.0.4 Server-Side Template Injection / Remote Code Execution
Python exploit framework for CVE-2026-4257 targeting vulnerable WordPress sites running Fluent Forms. The tool first detects a server-side template injection by submitting the Twig expression 77 to common form fields. If the target is vulnerable, it abuses Twig's registerUndefinedFilterCallback...
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...
