Forty scientists from 140 countries just told the United Nations something governments do not want to hear: the rules meant to control artificial intelligence are already behind, and the gap is widening every few months. The warning came in a preliminary report launched on Wednesday, July 1, by the UN Independent International Scientific Panel on Artificial Intelligence, and it lands five days before world governments sit down in Geneva to figure out what to do about it.

A Report Timed to Land Before Geneva

The panel, established by the UN General Assembly in 2025, is co-chaired by Turing Award winner Yoshua Bengio and journalist Maria Ressa, and its 40 members were selected from a pool of more than 2,600 candidates across 140 countries. UN Secretary-General António Guterres introduced the preliminary findings in a hybrid briefing on July 1, with Bengio and Ressa joining virtually to walk through the report ahead of the inaugural Global Dialogue on AI Governance, scheduled for July 6 and 7 in Geneva.

That timing is not an accident. The panel's mandate is scientific, not regulatory. It exists to hand governments an evidence base before they negotiate, rather than after. Whether the Geneva talks produce anything binding is a separate question, but the report itself is the clearest signal yet that the UN wants this conversation to start from hard numbers, not talking points.

The Compute Gap Driving Everything Else

Here is the number that should worry policymakers most: the United States controls roughly three-quarters of the computing power behind the world's leading AI supercomputers, and China accounts for around 15 percent, according to the report. Together, that is about 90 percent of the planet's frontier AI compute sitting in two countries. Most of the advanced models being trained on that hardware come from companies based in those same two places.

That concentration is not a footnote, it is the mechanism through which everything else in the report happens. Developing nations lack the infrastructure, technical talent, data, and local-language resources to build or even meaningfully audit the systems they are increasingly forced to depend on. The panel is blunt about where this leads: unless the gap closes, AI will widen global inequality rather than shrink it. A technology billed as a great equalizer is, on current trends, doing the opposite, and the report puts a number on why.

What the Panel Says AI Is Already Doing Right

To its credit, the panel does not frame AI as a threat in waiting. It documents what is already working. AI systems have predicted the structures of more than 200 million proteins, accelerating drug discovery, vaccine development, and antibiotic resistance research. Doctors are catching diseases like breast cancer earlier using AI-assisted screening, and health workers in developing countries are using local-language AI tools to improve patient care in places where specialists are scarce. Early-warning systems powered by AI are helping identify food insecurity before it becomes a full-blown crisis.

The panel's language here is deliberate: these are not future possibilities, they are already happening. That distinction matters because it undercuts the lazy framing of AI policy as a choice between innovation and caution. The report treats both the benefits and the harms as current facts, which makes the case for governance harder to dismiss as alarmist speculation.

The Risks the Panel Won't Soft-Pedal

The harms side of the ledger is just as concrete. The report names AI-generated sexual abuse material and sexually explicit deepfakes as an active threat, with women and children facing the highest risk. It flags disinformation convincing enough to erode trust in public debate and democratic institutions, and criminals already using AI for cyberattacks, fraud, and social engineering at scale. It also raises a harder-to-quantify danger: AI systems that reinforce harmful beliefs or behaviors in vulnerable users, contributing to mental health crises, including suicide.

On top of that, the panel points to a physical cost most conversations skip. AI agents can now plan tasks, use digital tools, and complete complex assignments with little human oversight, and researchers say the complexity of tasks these systems can handle has been doubling every few months. Meanwhile, the data centers running all of this consume enormous energy, adding to greenhouse gas emissions. AI's environmental footprint rarely gets equal billing next to its productivity gains, and this report refuses to let it disappear from the frame.

Why Existing Rules Aren't Working

More than 40 AI governance frameworks and ethical guidelines already exist worldwide, according to the panel, and that number sounds reassuring until you read the next line: they are fragmented, inconsistent, and rarely tested to see if they actually work. Worse, many of the safety assessments behind these frameworks are conducted by the same companies building the technology, which is a conflict of interest dressed up as due diligence.

The panel calls this the evidence dilemma. Regulators need solid data before writing rules, but by the time that data exists, the technology has already moved on. It is a genuine bind, not an excuse, and the report's proposed fix is straightforward on paper: independent evaluation, international cooperation, and shared standards that do not rely on self-grading. Getting governments and companies who profit from the status quo to actually adopt that is the much harder part, and the report stops short of explaining how that pressure gets applied.

What Happens Next in Geneva

The Global Dialogue on AI Governance opens Monday with this report as its evidentiary foundation. Member states will debate international approaches to managing AI, but the panel's own framing sets a low bar for success: the window for effective governance remains open, but it may not stay that way for long. That is not a deadline with a date attached, and it should have been. Vague urgency is easy for diplomats to nod along to and just as easy to ignore once the cameras leave Geneva. The real test of this report is not whether it gets cited in speeches this week, but whether any country walks out of that room with a binding commitment instead of another framework nobody enforces.