That 'Instant' ID Check Can Be Fooled by Words You Can't See

New security research demonstrates how invisible prompt injection attacks can hijack automated document verification pipelines, exposing a fundamental architectural flaw in how modern computer vision