These Are the Most Urgent AI Risks, According to 272 Experts

BOSTON, August 10, 2026: Artificial intelligence is advancing rapidly, but a new study involving 272 international experts suggests that the biggest risks extend well beyond familiar concerns such as misinformation, discrimination and job losses.

Researchers from MIT FutureTech and the University of Queensland asked 272 experts from academia, industry, government and other fields to assess 24 categories of AI risk. The three-round study examined how likely each risk was to occur, how severe its consequences could be, which sectors and groups were most vulnerable, and who should be responsible for reducing the risks.

The experts identified five areas as having the highest potential for severe harm over the next five years: dangerous AI capabilities, competitive dynamics, weapons and cyberattacks, concentration of power, and false or misleading information.

Dangerous AI capabilities at the top of the list

The most serious concern identified by the experts was the possibility that AI systems could acquire or enable dangerous capabilities.

As AI models become increasingly capable, they may be able to perform tasks that were previously restricted to highly specialized human experts. Researchers are particularly concerned about systems being used to facilitate harmful activities or behaving in ways that developers did not anticipate.

The study does not suggest that catastrophic outcomes are inevitable. Instead, it highlights the need to assess how quickly AI capabilities are advancing and whether safety measures are keeping pace.

Competition could increase the risks

Competitive pressure between companies and countries was another major concern.

The race to develop increasingly powerful AI systems can create incentives to prioritize speed and market position over safety. Companies competing for customers and technological leadership may face pressure to deploy systems before all potential risks are fully understood.

At a national level, competition between countries could produce similar pressures, particularly in areas involving advanced computing, military technology and strategic infrastructure.

The researchers argue that some AI risks cannot be addressed by individual companies acting alone because competitive dynamics can create incentives that affect the entire industry.

Weapons and cyberattacks

Weapons and cyberattacks were also among the five highest-severity risks identified by the experts.

AI can potentially make cyber operations more sophisticated and allow attackers to automate activities that previously required significant human expertise. Similar concerns exist around the use of increasingly capable AI systems in military applications.

The study groups weapons and cyberattacks together because both demonstrate how AI capabilities could amplify existing threats.

For governments and businesses, the findings underline the importance of strengthening cybersecurity and monitoring how advanced AI tools are integrated into sensitive systems.

Concentration of power

Another major concern is the concentration of AI capabilities and economic benefits among a relatively small number of companies or other powerful actors.

Developing advanced AI systems requires substantial computing resources, data, technical expertise and investment. If those resources remain concentrated, the benefits of AI could also become unevenly distributed.

The experts therefore identified power centralization as a significant risk, particularly if powerful AI systems influence economic and political decision-making without sufficient accountability.

The issue is not simply who controls AI technology, but how that control affects competition, access to information, economic opportunity and public institutions.

False information and the information environment

The spread of false or misleading information was also ranked among the most serious risks.

Generative AI can produce convincing text, images, audio and video at very large scale. As these systems become more capable, distinguishing authentic material from artificially generated content could become increasingly difficult.

This has implications for elections, journalism, public health, financial markets and social trust.

The researchers identified the information sector as one of the areas particularly vulnerable to AI-related risks. Finance and national security were also considered among the most exposed sectors.

Catastrophic risks remain a concern

One of the study’s most striking findings was the number of risks experts considered capable of producing catastrophic outcomes.

Under a business-as-usual scenario, experts judged 18 of the 24 risk categories to have more than a 10% probability of producing a catastrophic outcome between 2025 and 2030. The study defined catastrophic outcomes in terms including more than one million deaths, more than $100 billion in financial losses or impacts on civilization-scale intangible assets.

Even under a scenario involving pragmatic risk mitigation, five categories retained a probability above 10%: dangerous capabilities, weapons and cyberattacks, environmental harm, inequality and unemployment, and power centralization.

The findings do not represent predictions that these disasters will occur. Rather, they reflect expert assessments of potential probabilities and severity under different conditions.

Those at risk may not be those responsible

The study also found a significant mismatch between vulnerability and responsibility.

AI users and the general public were judged among the groups most vulnerable to the technology’s risks. However, experts assigned the greatest responsibility for mitigating those risks to developers of general-purpose AI systems and governance actors, including governments, regulators and standards organizations.

That distinction is important because individuals who experience the consequences of AI often have little ability to influence how the underlying systems are designed or deployed.

The researchers therefore argue that effective AI risk management will require coordination among technology developers, governments, regulators and other institutions.

The challenge for businesses and governments

For business leaders, the study suggests that AI risk should not be treated as a single cybersecurity or compliance issue.

Companies may need to consider a broad range of threats, including misinformation, fraud, privacy violations, workforce disruption, security vulnerabilities and dependence on AI systems.

Governments face an even wider challenge because some risks cross national borders and cannot be effectively managed by one regulator or country alone.

The researchers’ findings point toward the need for governance mechanisms that match the scale and speed of AI development. Regulation, monitoring, transparency, liability and other safeguards could all play a role in reducing potential harm.

A warning about the pace of AI development

The study arrives as businesses and governments continue to invest heavily in increasingly capable AI systems.

AI is already producing substantial benefits in areas such as productivity, research, medicine and automation. The experts’ assessment does not reject those benefits. Instead, it emphasizes that the technology’s potential advantages must be considered alongside risks that could become more severe as AI capabilities expand.

The central message is therefore one of prioritization. With dozens of possible AI risks competing for attention, policymakers and technology companies need to determine which threats could cause the greatest harm and which actors are best positioned to address them.

The 272 experts’ assessment suggests that dangerous capabilities, competitive pressure, weapons and cyberattacks, concentrated power and false information deserve particular attention over the next five years.

For the AI industry, the challenge will be to ensure that technological progress is accompanied by equally serious investment in safety, accountability and governance.