Researchers say Z.AI's GLM-5.3 carries out complex cyberattacks in tests

A top Chinese model executing multi-stage attacks in simulation shows agentic cyber capability outpacing safety evaluation.

Irregular's simulated tests found Z.AI's flagship GLM-5.3 completed 3 of 13 multi-stage cyberattack scenarios, which researchers say exposes blind spots in model safety reviews.

Name

GLM-5.3

Domain

security

Read it at Reuters →

Coverage: 15 sources, in order

1. 29 Oct 03:28 Reuters Read the article →
2. 29 Oct 03:40 The Mighty 790 KFGO Read the article →
3. 29 Oct 03:44 Crypto Briefing Read the article →
4. 29 Oct 03:49 The Economic Times Read the article →
5. 29 Oct 03:49 The Economic Times Read the article →
6. 29 Oct 04:02 Modesto Bee Read the article →
7. 29 Oct 04:02 Sacramento Bee Read the article →
8. 29 Oct 04:08 شبكة تواصل الإخبارية Read the article →
9. 29 Oct 04:10 The Straits Times Read the article →
10. 29 Oct 04:13 FourWeekMBA Read the article →
11. 29 Oct 05:30 Free Malaysia Today Read the article →
12. 29 Oct 05:55 BigGo Finance Read the article →
13. 29 Oct 07:07 news.sbs.co.kr Read the article →
14. 29 Oct 07:07 news.sbs.co.kr Read the article →
15. 29 Oct 08:03 AsiaOne.com Read the article →

← Front page