Chinese AI company Zhipu claims its new model is a better bug-finder than Anthropic, OpenAI
PLUS: HCL, TCS, admit data breaches; South Korea to fine Apple, Google; India bans some rideshare tips; and more!
Chinese AI company Zhipu has introduced a new AI model called GLM-5.3 that boasts bug-finding capabilities comparable to those of leading American models. The company's announcement highlights benchmark data indicating GLM-5.3 outperforms Fable 5 and GPT-5.6 Sol on the CyberGym benchmark, a test measuring a model's ability to address real-world cybersecurity challenges.
GLM-5.3 is classified as state-of-the-art on CyberGym for vulnerability discovery, and its performance significantly improves as it progresses through the exploitation chain, forming coherent plans for complete exploitation. The company has also collaborated with Chinese entities to test the model on actual codebases, uncovering 2,436 vulnerabilities across 269 projects, encompassing 1,097 medium-to-high severity issues spanning various software components.
Despite GLM-5.3's superior performance in bug-finding, it underperforms when compared to western models on other security and coding benchmarks. This superior bug-finding ability suggests China is rapidly closing the gap in cybersecurity expertise, potentially eroding the US advantage that previously stemmed from the home of AI models like Anthropic and OpenAI.
Written by urgent.news from The Register Science's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
