Tag
Epoch AI 研究发现，开源模型 GLM 5.3 能够利用安全补丁发布与用户更新之间的'patch gap'，自主构建针对 Firefox 的零点击文件窃取漏洞利用，成本仅约 1,000 美元/次，表明广泛可用的开源模型已能独立完成此前需要专家级能力的漏洞利用开发。
XBOW 拆解其六天内用 AI 智能体构建的 Chrome 全链漏洞利用：结合 V8 Maglev 编译器 bug、Animated WebP 解码器 bug 与 Crashpad 内存安全 bug，从普通网页 JavaScript 执行突破到渲染进程沙箱之外的代码执行，并获得 Google 首个 25 万美元 Chrome 全链 VRP 奖励。
Anthropic's report reveals that the Chinese open-weight AI model GLM-5.3 can build cyber exploits comparable to their Mythos Preview model, with easily bypassed safety guardrails, posing significant cybersecurity risks.
Eric Wallace announces the release of GPT-5.6-Cyber, a model focused on cybersecurity tasks like exploit development, and urges Sam Altman to consider using it for defensive purposes.
HackSys Extreme Vulnerable Driver (HEVD) is an intentionally vulnerable Windows kernel driver for learning kernel exploitation techniques, including use-after-free, stack overflows, and pool overflows. The tweet notes that a UAF exploit has been completed.
This repository provides artifacts for Segmentation Fault-Oriented Programming (SFOP), a novel exploitation technique that abuses signal handlers to bypass Intel CET. It includes PoC exploits and demonstrations against Nginx and Ladybird.
This article explores using large language models to assist in writing exploits for FreeBSD kernel vulnerabilities, detailing two exploit chains that achieve full jail escape.
This article describes how researchers at Praetorian used Claude Opus (via Claude Code) to discover and exploit vulnerabilities in the FreeBSD kernel, including a stack overflow (CVE-2026-3038) that allows escape from FreeBSD jails. Part one focuses on methodology for finding bugs.
This article from Anthropic evaluates how large language models like Claude Mythos Preview can accelerate the development of exploits for N-day vulnerabilities. Across tests on Firefox and Windows kernel patches, the model autonomously built working exploit chains, highlighting increased risks in the patch gap.
The author details the process of dumping, analyzing, and modifying HDD firmware to introduce a delay for an Xbox 360 exploit, covering hardware and software techniques without AI assistance.
Anthropic collaborates with researchers to benchmark Claude Mythos Preview on three new exploit development benchmarks (ExploitBench, ExploitGym, SCONE-bench), finding it outperforms all other models and demonstrating a step-change in LLM exploit capabilities.