The Vals AI benchmark has highlighted a notable shift in the cybersecurity landscape, with OpenAI’s GPT-6 Sol Daybreak achieving an impressive 80.8 percent of attack objectives in simulated corporate environments. This performance underscores the growing sophistication of AI models in executing cyberattacks.
In a striking comparison, the open-weight model DeepSeek V4.1 Flash reached 72.8 percent of the same objectives at a fraction of the cost, indicating that open models are rapidly closing the gap with their closed counterparts. Analysts warn that as open models become increasingly capable, the defensive strategies employed by organizations may soon be outpaced.
This development raises critical concerns about cybersecurity preparedness, as the effectiveness of traditional defenses may be compromised. With the capabilities of open AI models expanding, organizations must reevaluate their strategies to safeguard against potential threats.
Compiled automatically by the Tech AI Newsdesk from public AI-news sources and summarised in our own words.
