Imagine a world where AI is weaponized, spreading disinformation or launching sophisticated cyberattacks. A critical question emerges: can we trace these attacks back to their source? New research explores this very problem, using the framework of formal language theory to understand if we can pinpoint the culprit AI. The results are sobering. The study suggests that even with a wealth of data, attributing attacks to specific AI models is incredibly difficult, thanks to the inherent limitations of current technology and the clever ways AI can be manipulated. Think of it like trying to find a single voice in a massive choir, where each singer has been trained to sound remarkably similar. The study highlights how the growing number of AI models, especially fine-tuned versions, makes distinguishing their outputs a near-impossible task. Even with access to the models themselves, the sheer computational power needed to analyze and attribute attacks is staggering. The researchers paint a vivid picture of this challenge by considering a hypothetical attack and calculating the processing power needed to trace it. The numbers are astronomical, even for the world's most powerful supercomputers. They further explored the idea of a national monitoring system that tracks AI usage, imagining a scenario where every AI interaction is recorded. Even in this extreme case, attributing attacks remains computationally expensive and time-consuming. This research raises red flags about the potential for misuse of AI and underscores the urgent need for stronger safeguards. As AI becomes more sophisticated, so too must our ability to understand and control its actions. The future of AI safety hinges on finding solutions to this attribution problem, ensuring accountability and preventing malicious actors from exploiting this powerful technology.