Chinese AI Models Reach Advanced Cybersecurity Performance

Open-source artificial intelligence models are reaching technical capability levels comparable to the most restricted systems on the market. A study by the firm Vals evaluated the effectiveness of different technologies in simulations of attacks on complex corporate networks.

The experiment used two distinct scenarios to test machine autonomy. In one, the AI had to compromise a network to gain access to a second protected environment. The results challenge the idea that controlled access is the only way to contain the proliferation of advanced cyber skills.

OpenAI's Daybreak model performed best by completing 81% of the proposed tasks. However, the Chinese model DeepSeek achieved a 73% success rate. In several specific tests, the two tools recorded identical results.

This technological transition arrived on schedule, seven months after the forecast that such capabilities would be broadly available within 6 to 24 months. The availability of powerful models to any user changes the dynamics of global digital security. The cost to run these operations also becomes significantly lower with the use of open versions.

The findings suggest that the strategy of limiting access to frontier models may lose its practical effectiveness. The democratization of these capabilities requires immediate preparation by network defenders. The market must consider that sophisticated hacking tools are becoming widely accessible.