The advancement of autonomous AI systems has challenged the traditional foundations of criminal liability. This research examines whether direct criminal liability can be attributed to AI, relying on Iranian criminal law and concepts such as situational awareness, instrumental rationality, and conditional legal personality. The research was conducted using an analytical-critical method and based on library resources, Iranian laws, and comparative studies. The findings show that AI lacks human consciousness and will in the conventional sense, but some autonomous systems have functional situational awareness, the ability to predict and regulate behavior. These capabilities, although important for examining criminal attribution, are not equivalent to criminal intent or criminal liability alone. Based on the distinction between functional agency, normative competence, and criminal attribution, the article proposes a conditional and multi-layered model; in such a way that the possible liability of AI does not exclude the liability of developers, owners, users, and related legal entities. Also, responses appropriate to the technical nature of the system, such as modification, monitoring, restriction, quarantine, and disabling, are preferred over traditional punishments.