SAFE Project

SAFE Project Overview Overview of Activities in 2024 From Finalization to Publication
of Project Outcomes Related Materials & Database

SAFE Project Overview

  • - What is the SAFE Project?

    The SAFE (Safety and Assurance of Generative AI) project is an initiative to support the deployment of practical approaches to assure the safety of commercializing generative AI within GPAI. The Tokyo Expert Support Center positions supporting the SAFE project as its core mission.
    Please note that information about projects being implemented at the two centers established prior to Tokyo—Montreal (Canada) and Paris (France)—can be found on the following GPAI website:
    GPAI URL：https://oecd.ai/en/

  • - Background Leading to the Launch of SAFE

    In high-level government documents, including the “G7 Leaders’ Statement on the Hiroshima AI Process” agreed upon in October 2023, the G7 governments decided to advance project-based cooperation with GPAI and other organizations to ensure the safety of generative AI.
    Furthermore, in the Work Plan 2024 decided at the GPAI Summit held in India in December 2023, the launch of SAFE was explicitly stated as a new project based on the Hiroshima AI Process, and it was decided to promote the project along the following three steps.

  • - Three Steps

    The SAFE project is implemented in the following three steps:

    1. 1.IDENTIFY: Map all initiatives and companies focusing on AI safety solutions, with particular attention to related efforts and enterprises
    2. 2.ASSESS AND COORDINATE: Assess the effectiveness and feasibility of the identified initiatives and coordinate for collaboration
    3. 3.DEPLOY: Present solutions to GPAI member countries and international partners

Overview of Activities in 2024

  • - Establishment of the Tokyo Expert Support Center (July 2024)

    In July 2024, the GPAI Mid-Term Summit was held in India, where the establishment of the GPAI Tokyo Expert Support Center (hereinafter referred to as the “Tokyo Center”) was officially announced. At the same Summit, it was also decided that GPAI would form an integrated partnership with the OECD. As a result, the SAFE Project, for which the Tokyo Center is responsible, has been positioned as a project directly under the OECD.

  • - Establishment of a Framework under Co-Leads (July-September 2024)

    Inma Martinez and Cyrus Hodes, GPAI experts who had advocated the importance of SAFE since the end of 2023, were appointed as co-leads. Based on their direction, three thematic tracks were established: (1) Technical Safety of LLMs, (2) Data Governance of LLMs, and (3) AGI. Corresponding working groups were also created. Inma Martinez assumed leadership over Tracks (1) and (2), while Cyrus Hodes took charge of Track (3).

  • - From Project Launch to Interim Reporting (October–December 2024)

    The Tokyo Center introduced the three SAFE tracks and conducted an open call for participation (approximately 400 OECD and GPAI experts). As a result, 18 experts were selected. The first plenary meeting was held in October 2024, and subsequent activities were conducted by track-specific working groups. In total, seven working group meetings were held. In November 2024, project updates were shared at the GPAI Plenary, followed by further reporting at the GPAI Serbia Summit in December.

From Finalization to Publication of Project Outcomes

  • - Compilation of the Final Report (February 2025)

    In February 2025, a GPAI-related session was hosted by Inria during the Paris AI Action Summit. At this session, the Tokyo Center presented an overview of its activities and a summary of the SAFE Project.
    For Tracks (1) and (2), expert Kyoko Yoshinaga provided an evaluation of the current state of AI safety as of 2025, along with insights on the efforts of international networks such as national AISI organizations.
    For Track (3), Co-lead Cyrus Hodes emphasized the importance of mapping potential risks and solutions in the event of the development of AGI. As a key outcome, a searchable database was developed to allow easy access to both risks and corresponding mitigation strategies.

  • - Publication of Reports and Database (From April 2025 Onward)

    The SAFE Project’s report and the associated database are currently undergoing the necessary approval processes within the OECD and GPAI, with public release on the OECD.AI website and so on expected later in 2025.

  • - Initiatives for 2025

    The SAFE Project led by the Tokyo Center concluded its work at the end of 2024. Starting in 2025, the Tokyo Center will launch a new initiative on generative AI, based on the Workplan 2025 approved by GPAI member countries through the OECD. Details of this new project will be announced on the Tokyo Center’s website in due course.

Related Materials & Database

CONTACT Please feel free to contact us