www.mediafax.ro/social/modelele-ai-chinezesti-apro

Recent experiments indicate that publicly available AI models, particularly from China, are nearing the performance of restricted American systems in cybersecurity tasks. Vals, a company specializing in AI evaluation, tested various models' abilities to penetrate simulated computer networks. The results showed a significant reduction in the performance gap between advanced systems and those accessible to the public. OpenAI's GPT-6 Sol achieved the highest success rate, while the Chinese model DeepSeek performed impressively, completing a majority of tasks. The cost of using these AI tools for such operations varies greatly, with DeepSeek being much cheaper. This accessibility raises concerns about potential misuse, although it also aids security professionals in identifying vulnerabilities. The rapid development of AI models is automating tasks that previously required expert intervention, highlighting the evolving landscape of cybersecurity.