Z.ai's GLM-5.3 Challenges Anthropic's Mythos With Open-Source Security Innovations

Chinese AI startup Z.ai's open-source GLM-5.3 model has shown promise in cybersecurity, closely rivaling Anthropic's Mythos 5 in identifying software vulnerabilities. Despite some limitations, Z.ai's initiative aims to democratize access to advanced security tools via open-source development, sparking interest among international developers.

Chinese AI startup Z.ai has announced its open-source GLM-5.3 model, claiming a notable performance close to Anthropic's restricted Mythos 5 in identifying software vulnerabilities. The GLM-5.3 model achieved an 84.5% score on CyberGym, slightly surpassing Mythos 5's reported score, although independent verification is pending.

Despite lagging behind Mythos 5 in developing attack scenarios, Z.ai's GLM-5.3 remained a strong contender by completing 105 attack-development tasks in two hours and 130 in six hours. Mythos 5 managed 181 and 247 tasks, respectively. Z.ai plans to release GLM-5.3 publicly after enhancing its security assessments, with some functions reserved for verified users through a 'trusted access' program.

The company's move challenges the closed-access model of Mythos, advocating for open-source availability of advanced cybersecurity tools. Z.ai's launch includes the 'Open Source Shield' initiative to audit projects and provide defensive capabilities, echoing the ethos of Anthropic's 'Project Glasswing.' Interest in GLM-5.3 builds on its predecessor GLM-5.2's success, underscoring its potential in the cyber-defense domain internationally.