Roadmap and Strategy

We will proceed "over a time frame as short as realistically possible - yet with extreme care", as stated in our joint coalition announcement. While we set aspirational dates for our roadmap, our timeline will ultimately be governed by the time it will take for a critical mass of globally diverse states to join the initiative.

Our roadmap may have to be compressed or expanded significantly, depending on the evolving perception of AI advancements, of human safety and concentration of power risks and the timelines, and major AI-induced accidents.

While a slowdown in AI progress is possible, many AI scientists agree we may be a few years or one algorithmic innovation away from capabilities that will materialize those risks.

Waiting for a catastrophic AI accident to start building such global governance runs the risk of arriving too late to prevent loss of human control over AI, lead to a nuclear conflict, or to a durable, immense and undemocratic concentration of power in one or a few states or entities.

Setup Stage

  • By September 2024, we aim to have:

    • Launched a robust and clear website and established an essential social media presence.

    • Executed an email campaign to invite fitting NGOs, experts, volunteers, states and donors.

    • Started engaging states and regional IGOs, beginning with those showing interest in Q2 and Q3 2024.

    • Selected 3-5 leading, aligned, and globally-diverse NGOs to join the initial 6 as partners by signing the partnership agreement and agreeing to join the Coalition’s (Interim) Secretariat.

    • Selected several top experts for our high-level steering boards and working groups.

    • Secured some initial funding. 

    • Onboarded many globally diverse and reputable NGOs and experts as partners or members of the Coalition; speakers to our future Summits, Pre-Summits and other events; members of boards and working groups, and/or engaged in ad-hoc partnerships. 

    • On Sep 20-24, two partners and spokespersons of the Coalition attended the UN Summit of the Future in New York, engaging prospective states.

    • On Sept 26th, held a 2-hour presentation of the Coalition will be given to the Geneva Rotary Club International.

    • On Sept 27th, held a hybrid 2nd Pre-Summit in Geneva. 

First Stage

By November 2024:

  • The 1st Harnessing AI Risk Summit was held over two days in Geneva. 

  • Onboarded at least 3 to 7 states, that are globally diverse, via their participation to Summit or Pre-Summit as full participants or observers, and ideally:

    • committed to joining the next Summit and pre-summits and actively participating in State Working Groups, including elaborating an very initial provisional draft of the Mandate and Rules for the Election of the Open Transnational Constituent Assembly for AI and Digital Communications.

    • undersigned a LoI or MoU to join states.

  • Onboarded at least 30 experts and 20 NGOs, that are globally diverse and leading in AI safety, global governance or IT security, or former diplomats - via either:

    • participated in the Summit or other events as speaking participants and supporters of the Coalition.

    • joined our Working Groups, Advisory Board, as co-authors or contributors to white papers, academic papers and outreach materials to educate the general public and inform heads of states.

  • The coalition is more widely known and supported, possibly with mainstream media articles, interviews, podcasts, and videos of events.

  • States are becoming increasingly attracted to join the Coalition and Summits, understanding the scope and urgency of risk to human safety and concentration of power and wealth, the enormous economic advantages deriving from their co-leadership of the foreseen, and the potential political appeal of the Baruch Plan for AI for their voters.

By March 2025:

  • The 2nd Harnessing AI Risk Summit was held over 2 days in neutral state or participating state (or Geneva, or New York). 

  • Onboarded at least 7 to 9 states, that are globally diverse, including at least one medium-sized and one with rare or unique AI supply chain assets, and ideally:

    • are committed to join the next Summit and pre-summits, and actively participating in State Working Groups, including elaborating initial provisional drafts of the Mandate and Rules for the Election of the Open Transnational Constituent Assembly for AI and Digital Communications.

    • have undersigned a LoI or MoU to join states.

  • Onboarded at least 40 experts and 50 NGOs, who are globally diverse and leading in AI safety, global governance, IT security, and former diplomats.

  • States are increasingly engaged, developing a complete understanding of the AI situation, the risks and opportunities, and a sense of ownership of the initiative and getting down to the technical and treaty-making details.

  • Work is widely in place by Working Groups (ideally led overwhelmingly and equitably by states) to analyze in depth the socio-technical and geopolitical challenges of a successful global governance of the development and use of most powerful AIs, to play a role similar to that of the Acheson-Lilienthal Report, largely written by Oppenheimer, in providing the scientific basis for the proposal of the Baruch Plan in 1946. Such work is pursue via new documents and collaborations with leading NGOs, states and national security agencies on international initiatives - such as the International Scientific Report on the Safety of Advanced AI and the Guidelines for Secure AI System Development led by the NSA and GCHQ. 

By July 2025:

  • The 3rd Harnessing AI Risk Summit was held over 3 days in neutral state or participating state (or Geneva or New York). 

  • Onboarded at least 9 to 12 states, that are globally diverse, including at least one large, two medium-sized, and one with rare or unique AI assets.

    • A Caucus for the Coalition has been established at the UN General Assembly. 

    • Over one week of meetings a wide consensus has been built on comprehensive and detailed preliminary designs for the new intergovernmental organizations foreseen as outcome of such constituent assembly, that are used to inform the Mandate of the Assembly. 

  • Onboarded at least 50 experts and 100 NGOs, that are globally diverse and leading in AI safety, global governance or IT security, or former diplomats.

  • Widespread concerns, especially in the US and the UK, that creating new highly empowered global institutions inevitably leads to a concentration of power away from world citizens have been successfully countered by making clear that doing it right will result in the opposite.

By November 2025:

  • The 4th Harnessing AI Risk Summit was held over 4 days in neutral or participating states (Geneva or New York). 

  • Onboarded at least 15 to 30 states, highly globally diverse, that are globally diverse, including at least one large, two medium-sized, and one with rare or unique AI assets.

    • Drafts of the Mandate and Rules for the Election of the Open Transnational Constituent Assembly for AI and Digital Communications are details and well elaborate, nearing final version.

    • Superpowers and IGOs are also engaged in discussions.

  • Onboarded at least 60 experts and 200 NGOs, who are globally diverse and leading in AI safety, global governance, IT security, and former diplomats.

By March 2026:

  • the 5th Harnessing AI Risk Summit was held over 7 days in neutral or participating states (Geneva or New York). 

Second Stage

The Second Stage will start when - during a week-long Summit - at least 50 states - that highly globally diverse and make up at least 30% of the GDP and 30% of the world population, and include at least 2 states that are veto-holder of the UN Security Council or have nuclear weapons - have agreed during a week-long Summit on Mandate and Rules v.1 for the Assembly and convene the Assembly with a date, place and budget 6, 9 or 12 months later, and all states are invited to participate. 

Such Mandate and Rules should, we believe:

  • Be approved via simple majority based on weighted voting, after having extensively striven for consensus. Considering the vast disparity in power between states, particularly in AI and more broadly, and recognizing that three billion people are illiterate or lack internet access, we foresee the voting power in such assembly to be initially weighted by population and GDP.  

  • Have been widely perceived as fair, resilient, neutral, expert and participatory, including via transnational citizens' assemblies, and a neutral and balanced mix of experts.  

  • Ensure statutory safeguards are in place to maximize the chances that the resulting IGO will not degenerate, unduly centralize too much power or be captured by one or a few nations, and improve over time, such as via check and balances, a solid federal structure, regular mandatory re-constituent assemblies and other measures.

  • Ensure that the resulting charter is sent for ratification by the signing states, and becomes valid if 9/13th of them approve it, the same ratio used after the US Constitutional Convention of 1787.

  • Have established an open dialogue and convergence paths with competing or complementary international governance initiatives, especially those led by digital superpowers, and positioning itself as a medium-term complement for possible more urgent initiatives, albeit less multilateral, aimed to tackle urgent global safety risks.

  • By this time, or sooner, it is hoped that the US and China, other powerful states like Russia and Israel, and permanent members of the UN Security Council will all have joined. If China, the US, and most of those other states have joined by then, there can be enough positive and negative incentives exerted by participant states on non-joining states to keep their AI capabilities under reliable control (provided that AI supply chain control - and the public availability or hackability of dangerous AI technologies - have been mitigated early on via great powers coordination).

As decided, In 6, 9 or 12 months later, the Open Transnational Constituent Assembly for AI and Digital Communications is held in Geneva. Once it is clear the new organization will be created, we believe it will be participated in by at least 60 states, which are highly globally diverse and make up at least 50% of the GDP and 50% of the world population, and include at least 3 states that are veto-holder of the UN Security Council or have nuclear weapons. 

  • States non-participating to the Assembly can ratify as well.

  • Charter review provisions of the charter enable non-participating state to join in charter review conferences, initially held every 2 years, with equal rights.

  • Early ratifying states commit large enough funds to begin setting up the three agencies, including the Global Public Benefit AI Lab.