The UK government's AI Security Institute (AISI) has published an initial evaluation of the Mythos Preview model's cyber-attack capabilities. The study found that the model is not significantly different from other recent models in tests of cyber-security tasks, but could set itself apart from others due to its ability to chain tasks together into multi-step series of attacks. AISI also found that Mythos' ability to perform well in a 32-step data extraction attack on a corporate network is more impressive than other models.