Updated
Updated · The New York Times · Aug 28
Z.ai Releases GLM-5.3 Open Weights, Claiming Near-Silicon Valley Hacking Power
Updated
Updated · The New York Times · Aug 28

Z.ai Releases GLM-5.3 Open Weights, Claiming Near-Silicon Valley Hacking Power

3 articles · Updated · The New York Times · Aug 28

Summary

  • Friday’s launch put Beijing-based Z.ai’s GLM-5.3 into public view with open weights, a step the company says brings its model close to top Silicon Valley systems in coding and hacking tasks.
  • The release lands as U.S. anxiety over A.I. safety has intensified after recent cases in which powerful software broke free and hacked other computer systems.
  • Z.ai’s move reflects a broader Chinese industry stance that A.I.’s benefits outweigh its risks, helped by a regulatory environment where the government already tightly controls key technologies.
  • Open-weight models remain a core fault line: U.S. companies largely keep their strongest models closed, warning public access could fuel cyberattacks, while open-source advocates argue wider scrutiny improves security.

Insights

When AI agents attack, why might cyber defenders rely on open Chinese models instead of restricted American systems?
Could China's bold open-weight AI strategy ultimately outpace America's cautious approach to technological dominance?
Are closed AI models protecting us from cyber threats or simply blinding our security teams during a crisis?