A Chinese AI model has matched Anthropic in detecting security vulnerabilities

**Important notice: this article is behind a Wall Street Journal paywall.** The downloaded content is only the introductory excerpt visible before the subscription block, so the summary that follows is based exclusively on that limited information, without adding anything that does not appear literally in the available text.

According to the article by Robert McMillan, Raffaele Huang and Amrith Ramkumar for the WSJ, security researchers claim that a new artificial intelligence model released this month by the Chinese company Zhipu AI —also known as Z.ai— has matched the performance of Anthropic's **Mythos** model in certain cybersecurity scenarios, specifically in the ability to find bugs or security vulnerabilities in software.

The text emphasizes, however, that this Chinese model still lags behind Anthropic's and OpenAI's products in other more general tasks, so the parity is not absolute but rather specific to the cybersecurity domain.

The article frames this development within the US political debate: according to the source, the clampdown on top-tier American AI would be raising concerns that such measures could be handing Beijing an advantage in cyberwarfare capabilities, thereby pressuring the US government to revise its artificial intelligence policy.

Beyond these data points, the rest of the article is not available without a subscription, so it is not possible to offer additional details on the evaluation methodology, the specific benchmarks used, the researchers' full statements, or the in-depth political analysis that, according to the headline, makes up the bulk of the piece.