Rapid progress in artificial intelligence (AI) has raised concerns about the potential risks associated with its increasing capabilities and widespread deployment. At the same time, high uncertainty surrounding the feasibility and impact of these risks impedes effective policymaking. To help navigate this uncharted territory, we introduce the AI Risk Explorer (AIRE), an online platform monitoring the emergence of societal-scale AI risks with catastrophic potential.

In its initial phase, AIRE will focus on a narrow set of prominent risk scenarios, including AI-assisted biological attacks, AI-enabled cyberattacks, and loss of control over advanced AI. Awareness about those threat models has increased, and countries worldwide have committed to building a shared scientific understanding of that landscape. There is also an emerging corpus of evidence suggesting the risks might materialize. However, societal risk management is still hindered by several obstacles, such as technical barriers, time constraints, and information overload. Therefore, we need systematic efforts to gather data, distill the relevant information, and present it timely. AIRE is born to fit that purpose. By providing accessible information to policymakers and researchers, we aim to continuously improve their understanding of AI risk and enable more informed policy decisions. In particular, the project will highlight priority areas and hint at promising interventions while indicating which areas require more data and which practices should be further implemented to improve visibility over AI risk. 

AIRE will feature several components. First, a dashboard will collate relevant evidence on risk, such as frontier AI capabilities in relevant domains, demonstrations of those capabilities, and related incidents. Second, a policy tracker will display how governments address the risk through regulation and preparedness frameworks. The website will allow for an immersive experience where users can explore its multiple layers and progressively dive deeper into the topic according to their needs. Moreover, the website will also support a database where users can explore individual resources independently. At its basis, the content will be fed by an exhaustive data collection system.  

Ultimately, our vision is for AIRE to become an open-source intelligence hub capable of identifying early warning signs of risk emergence and catalyzing timely action. We will seek to work closely with external experts and public institutions to stay on track with the state of the art, understand the community’s needs, and support parallel risk assessment efforts. Join us in this endeavor and contribute to a safer future.