Frenos researchers tested six frontier models, two U.S. open-weight models, and two foreign open-weight models by providing them target details and asset information to assemble attack steps, without disclosing model names.