The LLM Attribution Crisis: Why Tracing AI Attacks is Becoming Impossible

Abstract: ‘Attributing AI-generated content to a specific model is becoming computationally intractable. We show that the hypothesis space for LLM attribution doubles approximately every 0.5 years due to the rapid proliferation of new models. Attributing just one year of U.S. AI-generated output would require an estimated 200 years of continuous supercomputing. This attribution crisis has significant implications for cybersecurity, disinformation detection, and AI governance, as it becomes increasingly impossible to reliably trace AI-generated attacks or synthetic content back to their source model.’

Use Google Scholar for full citation