How is the capacity to learn from mistakes a vital but poorly understood part of contemporary press freedom? If an autonomous press has the power to continually create and recreate the institutional arrangements that it thinks best enable public life, then this press must be able to manage its changes with knowledge and intention. It must be able to see, understand, revise, reject, and invent the relationships and systems that influence its reporting, writing, editorial judgments, publishing, and public missions. It must be able to know how and why its arrangements succeed or fail. Today, though, many of the press’s successes and failures intertwine with sociotechnical systems – platforms, algorithms, datasets, machine learning – that originate outside of newsrooms, have little appreciation or care for editorial nuance, and are simply hard to understand. How can a press be free if it lacks the full capacity to know these systems and, specifically, to know why these systems fail and when their failures matter to the press’s public function? This chapter aims to answer this question in three sections: first, tracing scholarship on how the press has historically defined and learned from its mistakes; second, analyzing cases when journalistic uses of generative artificial intelligence failed; and third, sketching a new press freedom framework that shows publishers and technologists alike the power of equipping journalists with the capacity to learn from sociotechnical mistakes.