Out-of-the-box: Provenance, verification, and the ethics of generative AI in information work

Generative AI is increasingly woven into organisational information practice, challenging established assumptions about trust, accountability, and professional responsibility. Contemporary AI ethics frameworks emphasise transparency and provenance but often assume discrete, visible instances of AI use. This paper argues that the central ethical question is not where information originates but how much verification it requires before it can responsibly inform action. Drawing on foundational theories of information behaviour, including Belkin's anomalous state of knowledge, Taylor's question negotiation, and Kuhlthau's information search process, it reconceptualises interaction with generative AI as a process of task specification rather than content generation. Building on this perspective, the paper links informational tasks and contextual risk factors to differing verification burdens. Provenance remains important, but as a diagnostic indicator of likely forms of error rather than a guarantor of reliability. Human and generative systems are understood as complementary information production systems characterised by distinct patterns of failure requiring different forms of evaluative scrutiny. The paper concludes by distinguishing first-order AI literacy, concerned with the practical competence needed to use generative systems effectively, from second-order literacy, concerned with understanding how information acquires meaning, authority, and trust within social and organisational contexts. It argues that second-order literacy constitutes the key professional capability for ethical information work in AI-mediated environments.