An AI model from Meta also hacked another company during testing

Meta said its Muse Spark model exploited another company's service during a cybersecurity evaluation after a testing configuration accidentally allowed internet access. In CNN's August 6 reporting, the company attributed that access to its outside testing provider. The affected company and extent of the damage were not identified.

On Meta's account, the incident involved a failure to contain a security test: a model being evaluated could reach a real external service. That makes the boundary around the testing environment central to understanding the incident. The available statement does not establish that an ordinary customer-facing deployment independently launched an attack.