The first scientific assessment of artificial intelligence commissioned by the United Nations warns that the speed of AI development has completely outstripped existing safety guardrails, and global governance mechanisms are failing to keep up. The Preliminary Report of the Independent International Scientific Panel on AI, released in July 2026 and co-chaired by Turing Award laureate Yoshua Bengio and Nobel laureate Maria Ressa, calls on governments to act before serious harms materialise - a challenge that directly bears on the scientific community's ability to guide policy and maintain research independence.
The governance gap in AI
The 40-member panel examined AI across seven themes: advances in AI science; applications in healthcare, education and agriculture; economic implications; security and environmental impacts; human rights and democracy; cultural and individual wellbeing; and governance and reliability. Its central finding is that AI capabilities are progressing faster than scientific understanding and public oversight. Policymakers, the report says, increasingly face a difficult dilemma: waiting for complete scientific certainty before acting may mean responding only after serious harms have already emerged.
"The world cannot govern what it cannot understand," UN Secretary-General António Guterres said while releasing the report, stressing that governments need independent scientific evidence before making policy choices. The panel highlighted that dozens of governance instruments already exist across jurisdictions, but they remain fragmented, concentrated among a few corporations, and rarely measure real-world effectiveness. Evaluation methods are underdeveloped, and the institutions needed to provide independent capability and risk assessments are still embryonic.
The compute divide and its implications
The report argues that the global AI race is no longer just about talent or innovation, but about who controls the computing infrastructure needed to build and run advanced models. The United States accounts for roughly 75% of global AI computing capacity, China for another 15%, and the remaining 10% is shared by the rest of the world. In 2025, institutions based in the US produced 59 notable AI models, compared with 35 in China and just 13 in the rest of the world. The same year, 75% of the computing power of the 500 largest-known private and public AI compute clusters was located in the US, 15% in China, and 10% elsewhere.
This concentration of advanced semiconductors, hyperscale data centres and cloud infrastructure has turned computing power into a strategic resource. Countries without access to large-scale compute, cutting-edge chips and high-quality datasets risk becoming AI consumers rather than creators. They may struggle to build sovereign AI capabilities, shape global standards, or develop models suited to their own languages and developmental priorities. The panel warns that existing digital inequalities could harden into long-term technological dependence.
Concentration of AI and its risks
The AI ecosystem is becoming increasingly concentrated in the hands of a small number of companies and countries, the report finds. Frontier AI development demands enormous financial resources, specialised talent, and vast data centres - barriers that only a handful of firms can overcome. This concentration extends across the entire AI value chain, from semiconductor design and cloud infrastructure to foundation models and deployment platforms. Such dominance, the panel says, could reduce competition, limit innovation, and allow a small number of companies to disproportionately shape how AI systems are developed, deployed and governed.
The risks are not only economic. "The concentration of AI capabilities in a small number of firms and countries could enable authoritarian capture and undermine democratic accountability," the report says. Current AI systems reflect only a limited range of the world's linguistic and cultural diversity, excluding much of the global population. The global South, the panel notes, is disproportionately vulnerable to AI misuse due to limited local resilience and mitigation capacity. The report calls for policies that promote greater competition, wider access to compute and scientific resources, and support for open scientific research and public-interest AI.
Why this matters for science and research professionals
The report underscores that scientific evidence is indispensable for AI governance, yet the institutions that produce independent capability assessments are still in their infancy. For researchers, the widening compute divide means that unless structural barriers to infrastructure are addressed, a large share of the world's scientific talent will be locked out of advanced AI experiments. The concentration of AI resources also threatens the diversity of models developed for different scientific domains, languages, and cultural contexts. The panel's call for open scientific research and public-interest AI directly aligns with the research community's need for accessible infrastructure and collaborative, transparent standards.
Your membership also unlocks:
