The first dimension, autonomy, describes how independently an AI system can operate. A low-autonomy assistant may generate a draft, recommend an action or wait for a human command before doing anything further. A more autonomous agent may break a broad instruction into subtasks, select tools, execute actions and adapt its behaviour based on changing conditions. At the highest levels, an agent could pursue a goal over extended periods with limited human intervention. Technically, this may involve planning loops, memory systems, tool access, environmental feedback and the ability to revise a sequence of actions. Each additional layer of independence changes the governance challenge, because oversight must move from checking individual outputs to monitoring an ongoing process.

Efficacy is the second dimension and refers to an agent’s ability to achieve intended outcomes in the world. A system can be highly autonomous yet ineffective, repeatedly taking actions without producing useful results. Conversely, an agent with modest independence may be extremely effective within a narrow domain, such as scheduling resources, analyzing specialized data or controlling a well-defined industrial process. Efficacy depends on more than language fluency. It can be shaped by access to reliable information, the quality of an agent’s planning and reasoning, its ability to use external software, the accuracy of its internal representations and the consequences of errors. As efficacy rises, failures may become less frequent—but successful actions can also create greater impact when the system is operating without close supervision.

The third dimension, goal complexity, concerns the structure of the objectives an AI agent is asked to pursue. Simple goals may involve a single, clearly defined task with an obvious completion condition. Complex goals can require an agent to balance competing priorities, interpret ambiguous instructions, coordinate multiple steps and respond to unexpected events. A request such as “find the best available option” may conceal questions about cost, reliability, fairness, timing, privacy and risk. For an AI system, translating such a request into operational decisions is not merely a matter of generating text. It requires selecting subgoals, assigning priorities and determining when a result is good enough. Governance must therefore address how goals are specified, who has authority to define them and how conflicts between objectives are resolved.

Generality forms the fourth axis. Narrow agents are designed for a limited task, environment or professional setting, where developers can test performance against a relatively stable set of conditions. General-purpose agents are intended to function across many domains, potentially moving between research, administration, programming, education and other activities. Greater generality can make systems more useful, but it also expands the range of situations in which their behaviour may be difficult to predict. A narrow system may be evaluated using domain-specific benchmarks and safety checks. A general system requires broader testing, because risks can emerge from combinations of capabilities that were not present in any single training example or evaluation scenario.

Together, the four dimensions produce what the authors call “agentic profiles.” Rather than asking whether a system is simply an AI agent, the framework asks what kind of agent it is and how its properties interact. A task-specific assistant might have low autonomy, moderate efficacy, simple goals and limited generality. An enterprise system that can access databases, coordinate workflows and act on behalf of employees could occupy a different profile, with higher efficacy and goal complexity. A highly autonomous, general-purpose agent would sit at the most demanding end of several axes, raising questions about supervision, accountability, security and control. The profiles are not intended as rigid labels; they are maps for comparing systems that may look similar on the surface but have very different real-world implications.

This approach also highlights why governance cannot rely on one universal safety measure. An autonomous agent with access to financial systems presents different concerns from a highly effective but closely supervised scientific assistant. A system with complex goals may require mechanisms for clarifying instructions, exposing intermediate plans and detecting conflicts between objectives. A general-purpose agent may need continuous monitoring across contexts rather than a one-time certification. Technical safeguards could include permission boundaries, sandboxed tool use, audit logs, human approval checkpoints, uncertainty reporting and emergency shutdown procedures. But the paper’s framework also points beyond engineering: organizations must determine who is responsible when an agent causes harm, how affected people can challenge decisions and what forms of deployment should require public oversight.

The researchers’ proposal arrives as AI developers increasingly describe their products as capable of acting rather than merely responding. Modern agents can combine large language models with retrieval systems, code execution, web browsing, APIs and persistent memory. In a typical control loop, the model receives an objective, observes information from its environment, generates a plan, invokes tools, evaluates results and repeats the process. This architecture can turn a probabilistic text generator into a system with practical influence over external events. The same loop that allows an agent to complete useful multi-step work can also amplify a mistaken assumption, an ambiguous instruction or a malicious input. Understanding an agent’s profile therefore becomes essential before deciding how much access, authority and autonomy it should receive.

The paper presents agentic profiles as a common language for developers, policymakers and the public at a moment when the boundaries between software and decision-maker are becoming increasingly blurred. The framework does not claim that every risk can be predicted from four dimensions, nor does it replace detailed technical evaluation. Its value lies in making differences visible: two systems may both be marketed as assistants, while one only suggests actions and the other executes them; two may be equally capable, while one operates in a narrow environment and the other generalizes across domains. By mapping autonomy, efficacy, goal complexity and generality, the authors argue, society can design governance mechanisms that match the actual power of an AI agent—before usefulness, scale and independence outpace accountability.

Subject of Research: AI agent capabilities and governance

Article Title: Agentic profiles for effective AI governance

Article References: Kasirzadeh, A., Gabriel, I. Agentic profiles for effective AI governance. Nature 656, 320–328 (2026). https://doi.org/10.1038/s41586-026-10805-z

Image Credits: AI Generated

DOI: 10.1038/s41586-026-10805-z

Keywords: artificial intelligence, AI agents, AI governance, autonomy, efficacy, goal complexity, generality, agentic profiles, AI safety, regulation

The study’s importance stems from the persistent tension in public discourse between sensational, far-future AI risk narratives and the very real, present-day problems that AI systems already cause. While existential risks – scenarios in which AI could hypothetically threaten humanity’s survival – often capture media attention and public imagination, they may inadvertently overshadow or distract from more pressing issues such as algorithmic bias, misinformation, privacy erosion, and labor market disruptions. The University of Zurich’s research offers the first large-scale, systematic data to empirically examine how framing AI risks influences public perception and concern.

To explore these dynamics, the research team undertook a series of three large-scale online experimental surveys involving more than 10,000 participants from the United States and the United Kingdom. These experimental designs exposed participants to varying narratives about AI: some read headlines emphasizing catastrophic, long-term threats to humanity; others encountered information focused exclusively on immediate concerns such as social discrimination, misinformation, and workforce impacts; while a control group viewed content highlighting the advantages and potential benefits of AI technologies. By comparing these groups’ responses, the researchers aimed to determine whether alarmist framing about existential risks diminishes public engagement with or awareness of more proximate problems.

A striking outcome from the experiments showed that individuals consistently expressed considerably greater concern about present-day AI-related harms compared to distant, theoretical catastrophes. Professor Fabrizio Gilardi of the University of Zurich’s Department of Political Science explains, “Our data highlight that people acknowledge the urgency of current risks like systemic bias encoded within AI algorithms and AI-driven job displacement. Even when confronted with apocalyptic warnings, their worry about these immediate problems remains substantially more pronounced.” This finding contradicts the criticism that emphasizing speculative future risks distracts from or undermines addressing urgent societal issues stemming from AI.

Notably, the study also reveals that people are capable of making nuanced distinctions between abstract, long-term existential threats and concrete, existing harms. Importantly, respondents did not exhibit a binary or mutually exclusive understanding of AI’s risks. Instead, they simultaneously recognized and took seriously the multifaceted challenges AI presents, both today and potentially tomorrow. This suggests that public concern should not be simplified or forced into an “either-or” framing when it comes to AI risk discourse.

The implications for the broader field of AI ethics, policy, and governance are significant. Historically, discussions of AI risk have often been polarized, with some arguing for urgent preventive measures solely against far-future scenarios, while others advocate focusing exclusively on mitigating current harms. The University of Zurich study encourages moving beyond this dichotomy by fostering a holistic approach that concurrently addresses immediate issues—such as bias amplification in automated decision-making systems and misinformation propagation—as well as invests in long-term research and safeguards against catastrophic risks.

This dual-perspective approach is especially crucial in times when AI technologies are increasingly embedded into key societal infrastructures. For example, facial recognition systems used in law enforcement can perpetuate racial and gender biases, leading to unjust outcomes. Social media algorithms powered by AI can exacerbate the spread of false information, influencing public opinion and democratic processes. Meanwhile, automation threatens to disrupt labor markets by making certain jobs obsolete, fueling economic and social uncertainty. Acknowledging these urgent challenges underscores the necessity of regulatory frameworks and oversight mechanisms tailored to present realities.

Moreover, the study’s nuanced findings provide valuable guidance for science communicators, policymakers, and AI developers aiming to engage the public effectively. Sensationalized warnings about an AI apocalypse, while attention-grabbing, should be balanced with clear communication about tangible harms and practical mitigation strategies to maintain public trust and foster constructive dialogue. As co-author Emma Hoes points out, “Public discourse should not be dominated by fear of the unknown but enriched with awareness of the concrete harms AI already causes, as well as the benefits it offers.”

Scientific literature had long speculated on the psychological effects of risk framing and media portrayal on public risk perception, but reliable empirical data in the context of AI risk narratives were scarce until now. This study fills that significant knowledge gap by demonstrating that existential risk narratives do not inherently suppress concern for immediate AI-related problems. Instead, awareness of both risk categories coexists robustly in the public mind, affirming the capacity for complex risk understanding among diverse populations.

The research team’s methodical approach, leveraging randomized exposure to different narrative framings, sets a new standard in studying perceptions around emerging technologies. This methodology can be applied to future investigations on risk communication strategies in other contentious technological domains. As debates unfold around AI-powered innovations like autonomous vehicles, medical diagnostics, and predictive policing, understanding how framing influences public caution and acceptance will be increasingly crucial.

While the study’s insights help to recalibrate the conversation on AI risk, they also call attention to the peril of complacency. The unmistakable fact that current AI risks command greater worry compels urgent policy responses and technological interventions. Regulatory bodies must prioritize transparency, accountability, and fairness in AI systems while ensuring that research into existential risks continues to receive due attention and funding.

In summary, the University of Zurich’s groundbreaking research reconciles two often competing perspectives within AI risk discourse. By establishing that neither present harms nor future dangers dominate or undermine the other in public concern, it underscores the importance of an integrative dialogue that encompasses the full spectrum of AI’s societal impact. Through precision in communication and policymaking, society can better navigate the complex landscape of artificial intelligence, harnessing its promise while safeguarding against its pitfalls.

Subject of Research: People

Article Title: Existential Risk Narratives About Artificial Intelligence Do Not Distract From Its Immediate Harms

News Publication Date: 17-Apr-2025

Web References: 10.1073/pnas.2419055122

References: Emma Hoes, Fabrizio Gilardi. Existential Risk Narratives About Artificial Intelligence Do Not Distract From Its Immediate Harms. Proceedings of the National Academy of Sciences. April 17, 2025.

Keywords: Artificial intelligence, Fear, Risk perception, Social psychology, Human social behavior, Society