GLM 5.3 autonomously exploits Firefox patch gap
先了解这件事
Epoch AI researchers demonstrated that the open-source model GLM 5.3 autonomously built exploit chains targeting Firefox vulnerabilities within 48 hours, successfully bypassing sandbox protections to read local files at a cost of approximately $1,000 per attack. The model analyzed public code changes to identify vulnerabilities in six Firefox versions released between September and October 2026, which Mozilla patched in version 159.0 after disclosure. The same approach failed against Chrome despite $100,000 in API costs, with researchers warning more advanced models could eventually overcome such defenses.
AI 根据报道生成 · 2 小时前更新
报道时间线
沿着报道，了解事件的不同侧面。
- The Decoder精选开源模型GLM 5.3在48小时内零人工干预构建完整Firefox攻击链
Epoch AI研究显示GLM 5.3能利用Firefox安全补丁发布时差自动构建攻击链，每次攻击成本约1000美元。该模型通过分析公开代码变更识别漏洞，成功绕过浏览器沙箱防护读取本地文件。研究团队向Mozilla披露后，相关漏洞已在Firefox 159.0中修复。测试中相同方法对Chrome无效，但研究者警告更先进模型可能突破防御。
本事件热度走势
还没有足够的连续观测数据，暂不绘制趋势。