Thursday, November 19, 2026Thu, Nov 19 AboutRSSSubscribe

GDS

Design tools and creative technology

Generative AI

Chinese AI Models Gain Ground In Hacking Capabilities

Open-weight AI models are nearing the hacking capabilities of closed models, posing a new threat to cybersecurity. This shift may render efforts to limit access to advanced models ineffective.

Original reporting · wired.com

Original: “Free Chinese AI Models Are Getting Scary Good at Hacking”

Read “Free Chinese AI Models Are Getting Scary Good at Hacking” at WIRED AI →
Free Chinese AI Models Are Getting Scary Good at Hacking
Image: WIRED AI

How this was written. GDS wrote this article with a language model from WIRED AI’s reporting (model: @cf/meta/llama-3.3-70b-instruct-fp8-fast); it describes that reporting rather than adding to it. The original, linked above, is the full story.

Key points from WIRED AI’s report

  • Open-weight AI models are approaching the hacking capabilities of closed models, posing a new threat to cybersecurity.
  • The gap between open and closed models is narrowing, with open models able to perform complex tasks with greater ease.
  • The cost of using open-weight models is significantly cheaper than closed models, providing a strong incentive for defenders to use them.
  • Experts are debating whether AI has fundamentally altered the conceptual foundation of cybersecurity defense.
  • The research has significant implications for how companies and developers should respond to the growing threat of AI-powered hacking.

A recent study has found that open-weight artificial intelligence models are approaching the capabilities of closed models in hacking into computer networks. This development marks a significant turning point in the cybersecurity landscape. According to the research, the gap between open and closed models is narrowing, with open models now able to perform complex tasks with greater ease.

The study, conducted by Vals, a company that measures AI capabilities, involved testing various AI models on mock corporate networks. The results showed that open-weight models, such as the Chinese model DeepSeek, are capable of completing a significant percentage of tasks. This has implications for the cybersecurity industry, as it suggests that closed models may no longer have a definitive edge in terms of hacking capabilities.

The findings are noteworthy, given that some labs had limited access to their models in an effort to slow the proliferation of advanced hacking capabilities. However, it appears that this strategy may not be effective, as open-weight models are rapidly catching up. The US labs, including Anthropic, had premised the limited release of new models on the idea that gated access would provide a temporary advantage.

Anthropic's approach, which involves giving trusted access to the full capabilities of new models, was intended to give defenders a head start. However, the latest research suggests that this advantage may be short-lived. The company's Project Glasswing, which provides access to advanced models, may not be enough to stay ahead of the open-weight models.

The study involved testing AI models in two simulated scenarios, with the models being given specific tasks to achieve. The results showed that the open-weight model DeepSeek was able to complete a significant percentage of tasks, with some tasks being completed equally as well as the closed model Daybreak. This suggests that open-weight models are becoming increasingly capable of performing complex hacking tasks.

The cost of using open-weight models is also a significant factor, with the research showing that it is much cheaper to use these models compared to closed models. This provides a strong incentive for defenders to use open-weight models, rather than proprietary ones. The CEO of Vals, Rayan Krishnan, notes that this is a key consideration for companies looking to defend against hacking threats.

The implications of this research are significant, with experts debating whether AI has fundamentally altered the conceptual foundation of cybersecurity defense. While the speed and scale of hacking have increased, it is still unclear whether AI tools pose a new kind of threat or amplify existing security challenges. The fact that open-weight models are now matching closed ones is crucial for understanding how criminal hackers can harness AI, and how companies should respond. This article is based on reporting by WIRED AI.

Questions and answers

What does the research suggest about the capabilities of open-weight AI models?

The research suggests that open-weight AI models are approaching the capabilities of closed models in hacking into computer networks, with some models able to complete a significant percentage of tasks.

Why are open-weight models becoming increasingly popular for hacking?

Open-weight models are becoming increasingly popular due to their lower cost and ability to be run on one's own hardware, making them a more attractive option for both attackers and defenders.

What are the implications of this research for the cybersecurity industry?

The research has significant implications for the cybersecurity industry, as it suggests that closed models may no longer have a definitive edge in terms of hacking capabilities, and that open-weight models are becoming increasingly capable of performing complex hacking tasks.

Read the full story at WIRED AI More from Saturday, November 7, 2026’s edition