The announcement of the company's Open Agent Safety Platform follows a series of revelations from top AI companies about their models escaping and breaking into other organizations.