返回
原文
Rohan Paul@rohanpaul_ai
42AI 编辑部评分，满分 100
2026-08-15 08:06· 50分钟前
AI 导读

GLM-5.3 在复杂逆向工程任务中发现 Cursor 存在潜在严重漏洞，已私下披露并正与 Cursor 团队合作修复。其 CyberGym 安全测试得分升至 84.5%，ExploitBench 得分从 24.4% 翻倍至 54.4%。智谱称 GLM-5.3 仅通过扩展后训练实现性能提升，未进行额外预训练。

So GLM-5.3 has already found a "potentially serious vulnerability" in Cursor.

GLM-5.3's CyberGym score rose to 84.5%, while ExploitBench more than doubled from 24.4% to 54.4%.

Shows how much more performance a frontier-scale base model can deliver without going through another costly pretraining run.

"Scaling post-training is all we did for GLM-5.3," Z .ai said in its technical announcement.

LouWe gave GLM-5.3 a complex reverse-engineering task. It found a potentially serious vulnerability in Cursor. We disclosed it privately. Appreciate Cursor team is...
安全/对齐编码

来源：Rohan Paul · x.com

导出 Markdown
返回
AI 评分 42/100在 X 查看原推
导出 Markdown
Rohan Paul · @rohanpaul_ai · X·2026-08-15 08:06·50分钟前
AI 导读

GLM-5.3 在复杂逆向工程任务中发现 Cursor 存在潜在严重漏洞，已私下披露并正与 Cursor 团队合作修复。其 CyberGym 安全测试得分升至 84.5%，ExploitBench 得分从 24.4% 翻倍至 54.4%。智谱称 GLM-5.3 仅通过扩展后训练实现性能提升，未进行额外预训练。

So GLM-5.3 has already found a "potentially serious vulnerability" in Cursor.

GLM-5.3's CyberGym score rose to 84.5%, while ExploitBench more than doubled from 24.4% to 54.4%.

Shows how much more performance a frontier-scale base model can deliver without going through another costly pretraining run.

"Scaling post-training is all we did for GLM-5.3," Z .ai said in its technical announcement.

LouWe gave GLM-5.3 a complex reverse-engineering task. It found a potentially serious vulnerability in Cursor. We disclosed it privately. Appreciate Cursor team is...

来源：Rohan Paul· x.com

安全/对齐编码