Anthropic warns Z.ai's open-weight GLM-5.3 poses cyberattack risk
Overview
Anthropic has released a report arguing that Z.ai's GLM-5.3, released in August and freely downloadable, poses a major cyberattack risk.
The report is part of a wider debate over whether frontier models should be open-weight, meaning anyone can download, retrain and study them, or kept behind closed APIs.
The Artificial Analysis LLM leaderboard shows closed models currently lead open ones in capability, with open models a handful of months behind. The source frames the question as whether the benefits of open models outweigh their risks, and does not settle it.
Written by AI from the articles below · updated Oct 13, 6:06 AM ET
Check the sources:
Article timeline
The articles in this story. Times are ET.
- MIT Technology Review · AINewsOpen models offer real benefits, but their risks raise the question of whether they outweigh them
AIAnthropic released a report arguing that Z.ai's GLM-5.3, released in August and freely downloadable, poses a major cyberattack risk. The debate centers on whether frontier models should be open-weight, which allows downloading, retraining and research, or kept behind closed APIs. Closed models currently lead open ones in capability, with open models a handful of months behind, according to the Artificial Analysis LLM leaderboard.
Heat trend
Not enough continuous observations to show a trend yet.