Executive Summary
Organizations continue to struggle with successfully integrating artificial intelligence into their operations, leading to cases where the technology fails to support, or even undermines, the organization’s mission. Researchers in academia, industry, and civil society have done tremendous work in developing techniques that can assist practitioners in their adoption of AI technology. Yet these contributions can easily be lost in the sea of collective AI efforts. While recent work to synthesize this research in key areas has begun, there remains no comprehensive guide to help practitioners navigate the maze of information. To address this gap, CSET researchers have collated over 1,200 resources from the breadth of academic papers, industry reports, and grey literature with the aim of making AI safety, security, and governance practices more accessible and easier to integrate into an organization’s existing processes. Building on CSET’s harmonized AI framework, this report collates actionable guidance from these resources into an easy-to-navigate format and connects the information back to core recommended practices, establishing a critical link between high-level principles and practical implementation details. This report aims to answer the core questions that practitioners need to ask to successfully operationalize AI best practices.
What: We identify the key techniques, tools, and practices that organizations should implement in adopting AI systems based on an extensive review of government guidelines, industry reports, academic papers, and the broader set of grey literature.
When: We map these implementation steps across the process of AI adoption, including planning, development, deployment, operation, and improvement.
Where: We provide guidance at all levels of the organization, ranging from organization-wide operations and practices to granular details at the data, model, and system level.
Who: We identify the functional teams within the organization that are most likely involved in carrying out each recommended implementation step.
How: We provide an extensive set of references alongside the implementation steps, identifying valuable resources from the staggering volume of available information.
Why: We link each step in our guide back to the recommendations from the harmonized framework that it serves to implement.
In addressing these key components, this report serves as a practical reference guide that bridges the gap between principles and practice for those adopting AI technology.