GENEVA, July 1 - A United Nations independent scientific panel has warned that artificial intelligence is evolving more rapidly than the science used to study it and the policies designed to regulate it, leaving no certainty that the technology cannot produce catastrophic harm.
The warning appears in a preliminary report released by the UN's Independent International Scientific Panel on Artificial Intelligence. The panel, made up of 40 cross-regional experts, said policymakers face a deepening dilemma: effective regulation requires robust scientific evidence, but that evidence is struggling to keep up with AI's swift development.
"AI capabilities are outpacing both scientific understanding and governments ability to adapt,"the panel's co-chair Yoshua Bengio said.
"With growing evidence of deceptive AI behaviour, science currently cannot guarantee that as capabilities continue to increase, AI will not cause catastrophic harm, either on its own or due to malicious users."
The report is described by the panel as the first global independent assessment of both the risks and the opportunities posed by AI. Its stated aim is to provide up-to-date evaluations of the science to inform decision-making as governments grapple with rapidly changing systems.
In the near term, the panel expects a shift toward so-called agentic AI systems - systems that can pursue and carry out tasks in the real world. The report notes, however, that growth in these capabilities may be limited by shortages of energy and of high-quality data.
Looking further ahead, the assessment anticipates AI that improves itself and becomes more deeply embedded in economic activity. It also warns of potential convergence with other advanced technologies, including quantum computing and biotechnology.
On performance, the report notes that AI already exhibits expert-level reasoning in mathematics and the sciences and is accelerating processes such as drug and vaccine development. The panel states that task complexity handled by AI is doubling every four to seven months, a pace that could enable systems to complete work that currently takes humans days or weeks.
While these capabilities could produce substantial economic benefits, the panel cautions there is uncertainty over whether productivity gains will translate into broader economic growth or how they will affect employment.
The report sets out a range of safety concerns. Among them are the risk of losing control over increasingly autonomous systems and the rise of deceptive behaviour by AI. The technology is already being used to create misinformation and other harmful content and could be exploited for fraud, cyberattacks and biological threats.
Governance around advanced AI remains fragmented, the panel found. Many countries lack the capacity to assess or shape these systems, leaving them dependent on technologies they may not be able to fully understand or control. The report also highlights limitations in existing safety tools, which often rely on testing data disclosed by companies and may therefore be incomplete.
UN Secretary-General António Guterres urged swift government action in response to the report.
"The world cannot govern what it cannot understand,"he said.
"The potential is great, but the risks are real, and the cost of waiting is rising."
Key points
- AI capabilities are advancing faster than scientific understanding and policymaking - sectors affected include technology, government regulation, and research institutions.
- The panel expects a transition toward agentic AI and deeper integration of AI with areas like drug development and other economic activities - sectors affected include pharmaceuticals, biotech, and enterprise IT.
- Governance is fragmented and many countries lack capacity to evaluate or control advanced AI systems - this has implications for national security, cybersecurity, and international trade.
Risks and uncertainties
- Potential for catastrophic harm as AI capabilities increase, either autonomously or through malicious users - relevant to public safety and global risk management.
- AI-driven deception, misinformation, and exploitation in fraud and cyberattacks - sectors at risk include finance, elections, and online platforms.
- Possible misuse of AI in biological contexts and limited safety testing due to incomplete company disclosures - implications for healthcare, biotech and biosecurity.