AI Accountability Gap: Who Controls the Power, Who Bears the Risk

Artificial intelligence (AI) is moving faster than the systems meant to govern it. What was recently a tool for answering questions and generating text can now write code, analyse vast datasets, create realistic images and videos, support scientific discovery and increasingly act with limited human supervision.

A preliminary report by the UN Independent International Scientific Panel on Artificial Intelligence states that the window to build effective global safeguards remains open, but may not stay open for long. The main concern is that its capabilities are expanding into areas where existing laws, institutions and oversight mechanisms are still fragmented, inconsistent or untested, making AI governance one of the defining policy tests of the next decade.

Governments are being asked to regulate systems they may not fully understand, at a speed traditional policymaking is poorly equipped to match. The result is a race between innovation and accountability, with high stakes for economies, democracies, workers, vulnerable communities and developing countries.

AI is shifting from assistant to actor

The debate around AI is changing because the technology itself is changing. The next wave of AI systems is not limited to responding to prompts. AI "agents" can increasingly plan tasks, use digital tools, write software and complete complex assignments with little or no human oversight. Researchers cited in the report say the complexity of tasks these systems can complete has been doubling every few months.

Autonomy changes the risk profile. The more capable and autonomous AI becomes, the harder it may be for regulators, users and even developers to understand how decisions are made, how errors spread and where responsibility lies.

The opportunity is still substantial. Used responsibly, AI could accelerate progress towards the Sustainable Development Goals (SDGs) by improving healthcare, education, scientific research, agriculture and accessibility for people with disabilities. It is already contributing to medical research, disease detection, local-language health support, food security early warning systems, personalised education and mental health support. However, the same qualities that make AI powerful also make it difficult to govern. Speed, scale and autonomy allow benefits to spread quickly, but they can also amplify harm before institutions have time to respond.

The benefits are global, but the power is concentrated

The report notes that the United States holds around three-quarters of the computing power behind the world's leading AI supercomputers, while China accounts for around 15 percent. Together, the two countries hold roughly 90 percent of that computing power. Most advanced AI models are also being developed by companies based in those two countries. This concentration has major implications. Computing power, data, technical expertise and investment are not just inputs into AI development. They shape who builds the systems, whose languages and social contexts are represented, who can audit the technology and who has leverage over how it is deployed.

Many developing countries lack the computing infrastructure, technical expertise, investment, data and local-language resources needed to fully benefit from AI. As a result, they may depend on systems they cannot build, inspect, audit or adapt to local needs. The risk is that AI could reinforce existing global inequalities rather than reduce them.

For low- and middle-income countries, the issue is not simply access to AI tools. It is sovereignty over digital development. If countries become consumers of externally built systems without the capacity to assess or adapt them, AI governance becomes not only a technology issue but also a development and geopolitical concern.

The risks are multiplying faster than oversight

The report identifies a broad set of risks that are already testing governments and societies. AI is being used to fuel online abuse, including sexual abuse material and sexually explicit deepfakes, with women and children most at risk. It can generate convincing false information, support cyberattacks, fraud and social engineering scams, reinforce harmful behaviours in some mental health contexts, and raise concerns about loss of control as systems become more autonomous. AI's energy-hungry data centres also add environmental pressure through greenhouse gas emissions.

These risks do not fall evenly. Women and children face particular exposure to image-based abuse. Citizens and democratic institutions face threats from disinformation. Workers may face disruption as AI systems take on more complex tasks. Governments must manage cybersecurity, accountability and public trust. Developing countries may face both exclusion from AI's benefits and exposure to its harms.

The governance problem is further complicated by what the report describes as an "evidence dilemma": policymakers need reliable scientific evidence before introducing regulation, but by the time enough evidence exists, the technology may already have moved on. More than 40 AI governance frameworks and ethical guidelines already exist, yet they remain fragmented and are rarely tested to see whether they work.

The gap leaves too much responsibility in the hands of the companies building the technology. Many safety assessments are conducted by developers themselves, underscoring the need for stronger independent evaluation, international cooperation and common standards.

Global rules must move from principles to proof

The UN is trying to build a more coordinated international architecture around AI. In 2025, the UN General Assembly established the Independent International Scientific Panel on Artificial Intelligence, made up of 40 experts from every region of the world serving in their personal capacity. Its role is scientific rather than regulatory: to assess the latest evidence on AI's opportunities, risks and impacts and provide independent reports for governments. The panel's work will feed into the UN Global Dialogue on AI Governance, which begins in Geneva on 6 July 2026. Member States are expected to discuss international approaches to managing the technology.

The test will be whether governments can move beyond broad ethical commitments toward systems that can be measured, enforced and trusted. It means independent safety evaluations, shared standards, transparency requirements, stronger accountability and investment in digital infrastructure and skills, especially for countries currently on the margins of AI development.

To sum up, the report states that AI is neither inherently good nor bad. Its impact will depend on the choices governments, companies and societies make now. The technology is already reshaping science, healthcare, education and economies. Whether it narrows inequality or deepens it, strengthens democracy or weakens trust, will depend on how quickly the world can build governance that keeps pace with innovation.