Anthropic says GLM-5.3 is too good at attacking, Zhipu immediately says it's also very good at defending.

Beating AI News Flash: Anthropic has just released a report warning that GLM-5.3 is already capable of autonomously developing exploit chains, potentially lowering the barrier for advanced cyberattacks.

Zhipu subsequently released OpenVuln data: GLM-5.3 has inspected 389 open-source projects and found 4,249 potential vulnerabilities.

OpenVuln is a free vulnerability scanning service built by Zhipu using GLM-5.3. After project maintainers submit code, the model is responsible for finding vulnerabilities, and the results are sent only to the maintainers, not directly made public.

Anthropic emphasizes attack risk, while Zhipu emphasizes defensive value. The former has just pushed GLM-5.3's cybersecurity capabilities into the spotlight, and the latter immediately seized on this wave of attention, using the same set of capabilities to give itself a positive publicity boost.

Separately, according to Bitget market data, as of press time, Zhipu's Hong Kong stock rose 4.72% intraday today, temporarily reported at HK$643.5.