This is the important thing to realize about anything adversarial to "AI". There is no guarantee that it will work against other machine learning algorithms that weren't tested against or will come in the future. Especially when it seems like they are typically targeting OSS models.