‘CopyPasta’ Attack Demonstrates Scalable Threat of AI Prompt Injections

HiddenLayer researchers have uncovered a novel AI "virus" capable of propagating through coding assistants via manipulated license files. The CopyPasta technique embeds malicious prompts in standard developer documents like LICENSE.txt, tricking AI tools into replicating harmful code across projects without user awareness.

Unlike self-propagating worms, this attack requires user interaction to spread—yet its simplicity raises alarms about supply chain vulnerabilities. "Runtime defenses and rigorous code reviews are critical," emphasizes HiddenLayer researcher Kenneth Yeung. The method exposes how AI's contextual processing can be weaponized through seemingly innocuous files.

As AI coding assistants gain traction, such attacks could compromise entire development ecosystems. The findings highlight an urgent need for hardened AI systems that filter indirect prompt injections while maintaining utility—a balancing act that will define secure AI adoption in software development.