AI News selected for Professionals and Decision Makers
Primary Research Stream

These are the most urgent AI risks, according to 272 experts

13:50 · July 20, 2026 · RSS APP - AI Primary Research

These are the most urgent AI risks, according to 272 experts

Which AI risks could cause the most harm in the next five years? New MIT research shows that businesses should be aware of threats like competitive pressure and dangerous AI capabilities.

Summary

MIT FutureTech and the University of Queensland used a Delphi process to gather structured judgments from 272 AI experts on 24 risk domains over a five-year horizon. The experts assessed both the likelihood and potential severity of harm under a business-as-usual scenario and under a pragmatic-mitigation scenario in which cost-effective safeguards are applied. Catastrophic outcomes were defined as events producing more than one million deaths, more than $100 billion in losses, or comparable civilizational damage.

Five domains stood out as carrying the highest combined likelihood and severity: dangerous capabilities that could enable persuasion, surveillance, or biological-weapon assistance; competitive pressures that encourage rapid deployment at the expense of safety; AI-enabled weapons and cyberattacks that exploit software vulnerabilities at scale; concentrated power arising from unequal access to advanced systems; and the generation and spread of false information. Even when experts assumed pragmatic mitigation measures, these five risks retained at least a 10 percent probability of catastrophic impact.

The information, national security, and finance sectors were judged most exposed. In information environments the primary concerns involve manipulation of content and erosion of trust; in national security the focus lies on offensive cyber operations and surveillance; in finance the risks center on fraud, market manipulation, and systemic instability. Across all three sectors, AI lowers the barrier for less-skilled actors while amplifying the speed and reach of already capable ones.

The study notes a persistent misalignment: AI developers and regulators are seen as bearing primary responsibility for risk reduction, yet the organizations and individuals most vulnerable to harm are often not positioned to influence those decisions. The authors therefore recommend that leaders embed AI-risk assessment into existing governance processes for cybersecurity, privacy, and business continuity rather than treating it as a separate compliance exercise.

Why it matters

This article provides empirical primary research on AI risk prioritization, which is highly actionable for Dutch researchers and enterprises aligning with the EU AI Act's risk-based approach. The focus on finance and information sectors directly applies to major Dutch industries seeking robust AI governance.

More in this beat
ai-alignmentbenefits-and-risksMIT FutureTechpolicy-and-societal-impactpolicy-implicationsrisk-and-limitations
Position: AI Lock-In Is in Progress, and We Must Be Prepared

06:00 · August 18, 2026

Position: AI Lock-In Is in Progress, and We Must Be Prepared

The article aligns strongly with the Dutch and EU focus on responsible, ethical AI and human oversight. Its proposed frameworks for mitigating systemic AI dependency offer actionable insights for Dutch policymakers, AI safety researchers, and enterprise leaders navigating AI adoption.

Relevance 85 · Audience 90

AI companions may worsen loneliness for vulnerable users

14:00 · August 4, 2026

AI companions may worsen loneliness for vulnerable users

This primary research is highly relevant for Dutch AI researchers and developers focusing on ethical AI and human-computer interaction. It provides empirical evidence on the psychological impacts of LLM-based companions, directly informing the design of safe, transparent, and socially responsible AI systems aligned with EU and Dutch ethical standards.

Relevance 75 · Audience 90

SysAdmin: Measuring Instrumental Power-Seeking in Frontier AI

06:00 · July 22, 2026

SysAdmin: Measuring Instrumental Power-Seeking in Frontier AI

This research is highly relevant for Dutch AI practitioners and researchers focusing on AI safety, ethics, and compliance with the EU AI Act. The SysAdmin benchmark provides an actionable framework for evaluating autonomous agents, which is critical for Dutch enterprises deploying AI in infrastructure and administrative roles.

Relevance 85 · Audience 95

Constructive Alignment: Governing Preference Dynamics in Human-AI Interaction

06:00 · July 2, 2026

Constructive Alignment: Governing Preference Dynamics in Human-AI Interaction

This research aligns perfectly with the Dutch AI market's strong strategic focus on ethical, transparent, and human-centric AI. It provides advanced researchers with a rigorous, control-theoretic framework to address the long-term societal impacts and potential manipulative risks of adaptive AI systems.

Relevance 85 · Audience 95

Position: The Term "Machine Unlearning" Is Overused in LLMs

06:00 · June 29, 2026

Position: The Term "Machine Unlearning" Is Overused in LLMs

This paper is highly relevant for Dutch AI researchers and compliance officers dealing with GDPR's 'right to be forgotten' and the EU AI Act. By clarifying the distinction between true machine unlearning and mere suppression, it provides a crucial framework for developing legally compliant and transparent LLMs.

Relevance 85 · Audience 95

Statement on the US government directive to suspend access to Fable 5 and Mythos 5

02:00 · June 12, 2026

Statement on the US government directive to suspend access to Fable 5 and Mythos 5

This is highly relevant for Dutch product teams and builders as it announces the immediate, global shutdown of Anthropic's Fable 5 and Mythos 5 models due to US export controls. Teams relying on these models must urgently migrate to alternative solutions to maintain their AI applications.

Relevance 95 · Audience 95

Toward Personal Intelligence Through Cooperative Observation

06:00 · August 19, 2026

Toward Personal Intelligence Through Cooperative Observation

Strong alignment with Dutch/EU priorities on ethical, transparent, and privacy-preserving AI; offers actionable concepts for researchers building user-owned personal agents compliant with GDPR and trustworthy AI guidelines.

Relevance 78 · Audience 85

Position: Evaluations of AI Moral Reasoning Still Miss Half of the Picture

06:00 · August 18, 2026

Position: Evaluations of AI Moral Reasoning Still Miss Half of the Picture

This article is highly relevant for Dutch AI researchers and practitioners focused on ethical AI, aligning strongly with the Netherlands' and EU's emphasis on transparent and trustworthy AI systems. It provides a critical framework for advancing LLM evaluation beyond simple value alignment toward robust normative reasoning.

Relevance 85 · Audience 95

AI Evaluation Should Work With Humans

06:00 · August 17, 2026

AI Evaluation Should Work With Humans

This paper aligns strongly with the Dutch and EU focus on ethical, human-centric AI and human oversight. It provides researchers with a conceptual foundation to develop new evaluation frameworks that prioritize human-AI collaboration over autonomous replacement, which is highly actionable for Dutch AI policy and enterprise deployment.

Relevance 85 · Audience 90

Stable Miscalibration in Large Language Models: A Practical View of High-Confidence Errors

06:00 · August 17, 2026

Stable Miscalibration in Large Language Models: A Practical View of High-Confidence Errors

This research is highly relevant for Dutch AI researchers and practitioners focused on trustworthy and transparent AI, a key priority in the Netherlands and the EU. Understanding stable miscalibration provides actionable insights for improving LLM reliability, auditing high-confidence hallucinations, and developing better abstention-aware decision policies for enterprise deployment.

Relevance 85 · Audience 95