“Sequent's research goal is to find principled methods to ensure AI alignment observed in controlled training environments generalizes to uncontrolled, real-world scenarios.”