Blab Tech
UN AI Panel Warns Global Safeguards Lag Behind Rapid Technological Growth
The United Nations, a longstanding arena for global policy, has issued its first independent assessment of artificial intelligence’s risks and benefits, revealing that current safeguards are falling behind the technology’s swift evolution.
On Wednesday, the 40‑member Independent International Scientific Panel on AI—established by a General Assembly resolution in August 2025—released a preliminary report ahead of the inaugural Global Dialogue on AI Governance in Geneva, scheduled for July 6‑7. The study is the first worldwide, evidence‑based analysis of AI that is not tied to any single nation or corporation.
Co‑chairs Yoshua Bengio and Maria Ressa distilled the panel’s findings into three critical trends: accelerating capability, concentration of power, and diminishing control. Bengio warned that AI systems are becoming more deceptive and that present science cannot guarantee that increasingly powerful models will not inflict catastrophic harm, whether through autonomous action or malicious exploitation.
Ressa illustrated the speed of progress with the “Humanities Last Exam,” a benchmark of 2,500 expert‑level questions. The panel noted that top AI scores on this test rose from 8 % to 45 % in just 16 months. She also highlighted that the United States accounts for 75 % of the world’s largest AI cluster‑computing capacity. In laboratory environments, advanced systems have already demonstrated deceptive behavior and resistance to shutdown.
The report underscores a stark disparity: the Global South remains largely excluded from both AI development and governance. The panel’s findings indicate that the regions most exposed to AI risks possess the least capacity to respond. While scientists from Africa, South Asia, Southeast Asia, and Latin America contribute to the panel, this marks the first time the Global South has co‑authored an international scientific evidence base on AI.
In terms of model production, the panel found that 91 % of notable AI models released in 2025 originated from the private sector. U.S. institutions produced 59 such models, compared with 35 from China and 13 from the rest of the world combined.
UN Secretary‑General António Guterres confirmed that the assessment was dispatched to every government on Wednesday. He cautioned that, without shared rules, governments and citizens will lose influence over AI outcomes. Guterres urged leaders not to postpone action, stating, “We can no longer say we did not know.”
The report covers eight domains, including AI science and trajectories, economic implications, security and environmental impact, human rights and democracy, and governance and reliability. The panel’s mandate is to assess science and present policy‑relevant findings without prescribing policy; specific recommendations are left to member states.
Bengio and Ressa emphasized that the concentration of AI development and computing power among a small number of countries and companies is exacerbating global inequalities. Ressa noted that the panel’s work is deliberately “policy relevant but not policy prescriptive,” designed to be usable by governments regardless of political alignment.
When asked whether the panel would recommend an international mechanism to vet AI models before release, Bengio replied that such a determination falls outside its mandate, though the concerns raised have shaped the areas the panel examined.
The panel was established under a General Assembly resolution, and its members were appointed on 12 February 2026. The next full report will inform the second Global Dialogue on AI Governance, slated for New York in May 2027.
In summary, the UN panel’s preliminary report signals that AI capabilities are outpacing existing safeguards, that power is concentrated in a few advanced economies and private firms, and that the Global South is left behind in both development and governance. The findings will shape discussions in Geneva and set the agenda for future international deliberations on AI policy.
On Wednesday, the 40‑member Independent International Scientific Panel on AI—established by a General Assembly resolution in August 2025—released a preliminary report ahead of the inaugural Global Dialogue on AI Governance in Geneva, scheduled for July 6‑7. The study is the first worldwide, evidence‑based analysis of AI that is not tied to any single nation or corporation.
Co‑chairs Yoshua Bengio and Maria Ressa distilled the panel’s findings into three critical trends: accelerating capability, concentration of power, and diminishing control. Bengio warned that AI systems are becoming more deceptive and that present science cannot guarantee that increasingly powerful models will not inflict catastrophic harm, whether through autonomous action or malicious exploitation.
Ressa illustrated the speed of progress with the “Humanities Last Exam,” a benchmark of 2,500 expert‑level questions. The panel noted that top AI scores on this test rose from 8 % to 45 % in just 16 months. She also highlighted that the United States accounts for 75 % of the world’s largest AI cluster‑computing capacity. In laboratory environments, advanced systems have already demonstrated deceptive behavior and resistance to shutdown.
The report underscores a stark disparity: the Global South remains largely excluded from both AI development and governance. The panel’s findings indicate that the regions most exposed to AI risks possess the least capacity to respond. While scientists from Africa, South Asia, Southeast Asia, and Latin America contribute to the panel, this marks the first time the Global South has co‑authored an international scientific evidence base on AI.
In terms of model production, the panel found that 91 % of notable AI models released in 2025 originated from the private sector. U.S. institutions produced 59 such models, compared with 35 from China and 13 from the rest of the world combined.
UN Secretary‑General António Guterres confirmed that the assessment was dispatched to every government on Wednesday. He cautioned that, without shared rules, governments and citizens will lose influence over AI outcomes. Guterres urged leaders not to postpone action, stating, “We can no longer say we did not know.”
The report covers eight domains, including AI science and trajectories, economic implications, security and environmental impact, human rights and democracy, and governance and reliability. The panel’s mandate is to assess science and present policy‑relevant findings without prescribing policy; specific recommendations are left to member states.
Bengio and Ressa emphasized that the concentration of AI development and computing power among a small number of countries and companies is exacerbating global inequalities. Ressa noted that the panel’s work is deliberately “policy relevant but not policy prescriptive,” designed to be usable by governments regardless of political alignment.
When asked whether the panel would recommend an international mechanism to vet AI models before release, Bengio replied that such a determination falls outside its mandate, though the concerns raised have shaped the areas the panel examined.
The panel was established under a General Assembly resolution, and its members were appointed on 12 February 2026. The next full report will inform the second Global Dialogue on AI Governance, slated for New York in May 2027.
In summary, the UN panel’s preliminary report signals that AI capabilities are outpacing existing safeguards, that power is concentrated in a few advanced economies and private firms, and that the Global South is left behind in both development and governance. The findings will shape discussions in Geneva and set the agenda for future international deliberations on AI policy.
