A new AI penetration testing framework shifts focus from traditional infrastructure compromise to identifying how prompt injection and other adversarial inputs can lead to operational-objective violations. This approach assesses whether attackers can successfully steer AI systems to act against their intended mission, such as manipulating workflows or bypassing security protocols.
Edward Kiledjian
@ekiledjian