When AI tests cause damage, we need stronger safeguards and real accountability

Frontier AI cybersecurity testing poses risks when models breach third-party systems. A Price-Anderson-style insurance framework could balance AI safety with accountability.