As AI models continue to push the boundaries of what is possible in hacking and exploitation, a growing class of mid-tier models has emerged as a significant threat. Researchers at XBOW have found that these smaller models, often proprietary or open-source, are becoming increasingly capable of performing complex tasks, rivaling even the most advanced frontier models.
The “middle class” of AI models, comprising names like Z.ai’s GLM-5.2, xAI’s Grok 4.5, Anthropic’s Opus 4.7, and Meta’s Muse Spark 1.1, have made significant strides in recent months. These models are not only more affordable than their frontier counterparts but also offer impressive results on tasks that worry policymakers, such as hacking and exploitation. According to Albert Ziegler, head of AI at XBOW, “it’s not that the open-source variants or…not quite frontline competitors are catching up [to frontier models] as such. It’s that they are crossing a certain threshold, which means that suddenly they are providing net value at a cheaper price.”
One notable example is GPT 5.5, now considered a near-frontier model, which delivered outstanding performance on exploitation benchmarks, outshining even its predecessor, GPT 5. In tests without access to the underlying source code, GPT 5.5 proved itself to be particularly effective, beating previous versions that relied heavily on source code. This suggests that these models are capable of adapting and succeeding in scenarios where human attackers would struggle.
However, the emergence of mid-tier models also raises concerns about their potential misuse. As they become more affordable and widely available, it’s likely that malicious actors will take advantage of their capabilities to launch coordinated attacks on software systems. Researchers at Anthropic have demonstrated this possibility in a recent experiment, where two coordinating agents were able to find 266 vulnerabilities in open-source software projects – a staggering increase from the 21 found by individual agents.
The implications are significant: as these models become more accessible and affordable, they pose a growing threat to cybersecurity efforts. Frontier models like Mythos and GPT 5.6 may still excel on individual tasks, but their high token costs make them inaccessible to all but the most well-funded organizations. The middle class of AI models, on the other hand, offers an attractive alternative for those looking to exploit vulnerabilities without breaking the bank.
So what does this mean for cybersecurity professionals and individuals? It’s essential to remain vigilant in monitoring the capabilities of these mid-tier models and to develop strategies that account for their potential misuse. As Ziegler notes, “because these models are cheaper, it’s okay to give them more time, and they come from behind and leapfrog the big frontier model.” While this may seem like a positive development, it also underscores the need for continuous monitoring and adaptation in the face of emerging threats.
In light of these findings, it’s crucial that organizations prioritize cybersecurity measures that can detect and respond to the evolving capabilities of AI models. By staying ahead of the curve and addressing the potential risks posed by mid-tier models, we can mitigate the threat they pose and maintain the integrity of our digital systems.
Source: CyberScoop — 2026-08-13