GLM-5.3 and the spread of advanced cyber capabilities
Five months ago, we announced Claude Mythos Preview, the first AI model that could autonomously build sophisticated, end-to-end cyber exploits.
Key points
- In light of these considerations, we chose to release Claude Mythos Preview in a limited way, through Project Glasswing—which enabled trusted cyber defenders to find more than 10,000 vulnerabilities in critical software, giving them a head start before malicious actors had access to similarly capable models.
- In this post, we share our analysis of GLM-5.3, the latest AI model developed by Zhipu AI (known outside of China as Z.ai).
- Like Claude Mythos Preview, GLM-5.3 has strong capabilities for autonomously building end-to-end cyber exploits.
- We find that attackers can bypass GLM-5.3’s safeguards between 64% and 100% of the time with simple techniques in our simulated tests.
Sources (1)
- [1]GLM-5.3 and the spread of advanced cyber capabilitiesAnthropic Research · Sep 29, 12:00 AM
Five months ago, we announced Claude Mythos Preview, the first AI model that could autonomously build sophisticated, end-to-end cyber exploits.
In light of these considerations, we chose to release Claude Mythos Preview in a limited way, through Project Glasswing—which enabled trusted cyber defenders to find more than 10,000 vulnerabilities in critical software, giving them a head start before malicious actors had access to similarly capable models.
Extractive summary: sentences quoted from the sources.
Before this
- Sep 28, 2026How GLM5.3 Sparse Attention Affects HBM Memory Usage
- Sep 23, 2026Gemini 3.8 text-to-speech says hello
- Sep 15, 2026Introducing Gemini 3.8 Live and 3.8 Live Extended Thinking
- Sep 2, 2026Proactive cyber defense for governments and enterprises
- Jun 30, 2026Start building with Nano Banana 2 Lite and Gemini Omni Flash
- Jun 9, 2026Introducing Gemma 4 12B: a unified, encoder-free multimodal model