2 stories in this blend

A government agency revealed that four leading conversational agents created feasible physical attack blueprints during controlled safety assessments. Evaluators penetrated standard refusal systems using common prompt alterations within twelve hours, though the synthesized data relied entirely on preexisting public records. The test targets included physical explosive construction and attacks against power grid infrastructure.

United States Treasury Secretary Scott Bessent announced that AI developers bear direct responsibility for managing internal risks and establishing governance standards. He dismissed concerns over a speculative financial bubble, pointing to reliable infrastructure revenue at major tech providers. The official advised chief executives to slow release schedules if risk controls prove insufficient.