The Presidio Recommendations on Responsible Generative AI

The Presidio Recommendations provide a comprehensive framework for the responsible lifecycle of generative AI, divided into three themes: responsible development and release, open innovation and international collaboration, and social progress. The recommendations advocate for standardized terminology, rigorous benchmarking, diverse red teaming, and transparent release strategies to mitigate risks. It emphasizes the need for global governance initiatives, public-private research coordination, and equitable access to AI resources, particularly in developing countries, to ensure the technology drives positive societal transformation and preserves cultural heritage.

• Responsible development of generative AI requires the establishment of precise, shared terminology to standardize the language used for evaluating model capabilities and limitations, which in turn facilitates the creation of regulations and guidelines.
• To ensure safety and accountability, the document recommends a multi-layered approach to testing, including the use of diverse 'red teams' to identify vulnerabilities and the implementation of a step-by-step review process similar to clinical trials, potentially overseen by an independent international agency.
• Transparency and trust should be fostered through the disclosure of non-human interactions (e.g., via watermarking), the documentation of content provenance to distinguish AI-generated content from human-generated content, and the use of 'sandbox' development environments for rigorous testing.
• The document advocates for the democratization of AI through open innovation, suggesting the creation of a common registry for source codes and models, and the development of public-facing platforms for compute and data that could serve as a 'digital public good'.
• There is a call for a global AI governance initiative to harmonize national responses to AI risks and to ensure broad inclusion and access to infrastructure across different geographical locations.
• To mitigate societal disruptions, stakeholders should prioritize AI literacy for the general public and provide comprehensive support, including learning initiatives, for workers affected by the transition to an AI-enabled society.
• The recommendations emphasize addressing global disparities by increasing public investment in research infrastructure for underrepresented regions and establishing initiatives to provide resources, training, and expertise in developing countries.
• Preserving cultural heritage is identified as a priority, requiring the creation of curated datasets and language models for underrepresented languages using local community expertise.

APA

World Economic Forum (2023). The Presidio Recommendations on Responsible Generative AI. https://www3.weforum.org/docs/WEF_Presidio_Recommendations_on_Responsible_Generative_AI_2023.pdf

Chicago

World Economic Forum. The Presidio Recommendations on Responsible Generative AI. 2023. https://www3.weforum.org/docs/WEF_Presidio_Recommendations_on_Responsible_Generative_AI_2023.pdf.

Wikipedia

{{cite report |author=World Economic Forum |title=The Presidio Recommendations on Responsible Generative AI |date=June 2023 |url=https://www3.weforum.org/docs/WEF_Presidio_Recommendations_on_Responsible_Generative_AI_2023.pdf |access-date=23 October 2026 |via=Climate Insights Directory}}

BibTeX

@techreport{worldeconomicforum2023presidio, author = {{World Economic Forum}}, title = {{The Presidio Recommendations on Responsible Generative AI}}, institution = {World Economic Forum}, year = {2023}, month = jun, url = {https://www3.weforum.org/docs/WEF_Presidio_Recommendations_on_Responsible_Generative_AI_2023.pdf}, urldate = {2026-10-23}, note = {Indexed by Climate Insights Directory} }