UN Launches First Global Dialogue on AI Governance in Geneva with 193 Countries at the Table
Meeting at Palexpo, with 193 member states co-chaired by El Salvador and Estonia, aims to establish a multilateral floor as the US, European Union, and China pursue divergent paths.
Palexpo in Geneva is hosting the opening session of the UN's first Global Dialogue on AI Governance on Monday, July 6. Convened by the General Assembly, this gathering unites all 193 member states for the first time, alongside companies, civil society, and the technical community, under the co-leadership of Salvadoran diplomat Egriselda López and Estonian Rein Tammsaar. The agenda continues until July 7 and seamlessly transitions to the AI for Good Global Summit by the ITU, from July 7 to 10.
Secretary-General António Guterres summarized the goal in a sentence that sets the tone for the week: "The question is no longer whether AI will transform the world. It already is. The question is whether we will govern this transformation together or let it govern us." Guterres structured his speech around three pillars: policy, science, and capacity, and outlined three operational goals: safe and overseen systems, interoperability among governance regimes, and open innovation with accessible open-source tools for countries outside the technological frontier.
What Science Tells Diplomats
The day's agenda includes a plenary presentation of the preliminary report from the International Scientific Panel on AI, released on July 1. The panel, composed of 40 experts, is co-chaired by Turing Award winner Yoshua Bengio, scientific director of Mila in Montreal, and journalist Maria Ressa, 2021 Nobel Peace Prize laureate. The document states directly that "science currently cannot guarantee that as capabilities continue to expand, AI will not cause catastrophic harm, either on its own or through malicious use." This statement will be echoed by delegations advocating for binding rules and will counterbalance the discourse from laboratories calling for self-regulation.
The panel also signaled something more uncomfortable for compliance areas: there are currently no technical guarantees that agent systems will consistently follow the instructions received. For CIOs and CISOs already operating pilots with agents in production, the message is operational, not philosophical.
Three Blocs, Three Paths
The Dialogue arrives at a time when the main regulatory blocs are moving in distinct directions. In the United States, Donald Trump signed an executive order on June 2 called Promoting Advanced Artificial Intelligence Innovation and Security, which requests companies to voluntarily submit frontier models for federal review within 30 days prior to public launch, a shortened period from the initial 90 days considered. It is an advisory regime, not licensed.
In the European Union, the Council gave final approval on June 29 to the Digital Omnibus VII package, which adjusts the timeline for the AI Act but maintains the key date: the full application of core obligations is stalled for August 2, three days after the anticipated implementation of the simplified text. Providers of general-purpose models come under direct supervision of the Commission. China, which at the Beijing Summit in December tried to push its own multilateral framework, arrives in Geneva under pressure to explain unilateral restrictions on access to models, such as the recent blockade by Anthropic against subsidiaries in Singapore.
In the field of global service delivery, the Dialogue matters through another avenue. India, the Philippines, and Poland, offshoring hubs employing millions, are highlighted in the report as the countries most vulnerable to AI's "productivity deflation" in outsourcing contracts. Kotak estimates an average discount of 3% to 3.5% already embedded in new managed service contracts. A governance regime that fragments labeling, auditing, and civil liability over autonomous agents directly affects compliance costs for suppliers in these regions and the design of SLAs with global clients.
The Signal for the C-Level
The Dialogue will not produce binding text. What emerges from Geneva this week is a draft of common language, and it is this language that will appear in contracts, risk frameworks, and due diligences over the next 24 months. For those purchasing or deploying frontier models, three documents gain immediate weight: the scientific panel's report, the consolidated text of the AI Act post-Omnibus, and the list of 44 members of the AI for Good Global Commission, which meets for the first time on July 8 at Palexpo and includes CEOs from frontier labs alongside heads of state. This marks the first time Jensen Huang, Andy Jassy, and Jack Clark formally sit in a governance body under a UN mandate. The European advisory model, the American voluntary approach, and the Chinese dirigiste will need to coexist with this new architecture, and the global C-level will have to decide which of these it wants to be closer to.