In a dramatic shift within the AI security landscape, open-weight models have rapidly advanced to match the cyber capabilities of closed frontier models from just four months ago—while costing a fraction of the resources, according to a new report from the British AI Security Institute (BAISI).
The report highlights a significant narrowing of the performance gap between open and closed models in the realm of cybersecurity. At the beginning of 2025, open models lagged by six to ten months. By mid-year, that gap had shrunk to just four to seven months, signaling a rapid evolution in open-source AI capabilities.
Implications for Cybersecurity Defenders
BAISI’s findings raise concerns about the growing risks posed by open-weight models. Despite their improved performance, the institute found that many safety measures implemented in these models are largely ineffective. This leaves cybersecurity professionals with less time to adapt and defend against potential threats, as the models become more capable and accessible.
"The rapid advancement of open-weight models is a double-edged sword," said a spokesperson from BAISI. "While they offer cost-effective solutions, their increasing sophistication presents new challenges for defenders who must now prepare for threats that were previously out of reach."
What This Means for the Future
The findings underscore a critical need for updated cybersecurity frameworks that can keep pace with the evolving AI landscape. As open-source models continue to close the performance gap with their closed counterparts, the industry must balance innovation with safety. Organizations relying on AI for defense will need to reassess their strategies and invest in adaptive technologies to stay ahead.
With open models now offering near-frontier performance, the question isn’t whether they’ll become more powerful, but how quickly they’ll be weaponized and how well defenders can respond.
