Automated framework for frontier AI safety evaluation that synthesizes executable, pressure-rich test environments to uncover latent risks in advanced agent models.