Source: Getty
research

Infrastructure for AI Agents

The world needs external protocols and systems that shape how agents interact with institutions and other actors.

by Seth Lazar, Alan Chan, Kevin Wei, Sihao Huang, Nitarshan Rajkumar, Elija Perrier, Gillian K. Hadfield, and Markus Anderljung
published by
Transactions in Machine Learning Research
 on May 10, 2025
Read the Publication
AITechnology
Carnegie does not take institutional positions on public policy issues; the views represented herein are those of the author(s) and do not necessarily reflect the views of Carnegie, its staff, or its trustees.

More work from Carnegie

  • Flags representing some of the 30 different nations taking part in the 15th African Union Summit flutter at the entrance of the convention hall on the second day of the meeting in Kampala, Uganda on July 26, 2010.
    U.S. Capitol Building at twilight
    Abstract image for artificial intelligence
    collection
    Artificial Intelligence

    As artificial intelligence (AI) changes how people around the world live and work, new frontiers for international collaboration, competition, and conflict are opening. AI can, for example, improve (or detract) from international cyber stability, optimize (or bias) cloud-based services, or guide the targeting of biotechnology toward great discoveries (or terrible abuses). Carnegie partners with governments, industry, academia, and civil society to anticipate and mitigate the international security challenges from AI. By confronting both the short-term (2-5 years) and medium-term (5-10 years) challenges, we hope to mitigate the most urgent risks of AI while laying the groundwork for addressing its slower and subtler effects.

  • AI and California
    paper
    Carnegie California AI Survey

    Californians believe that AI will significantly impact their work, communities, and democracy. But this is offset by high levels of anxiety and uncertainty around specific impacts.

    • +3
  • U.S. Capitol Building at twilight
    article
    When Should Congress Preempt State AI Law? The Lessons of Past Technologies

    Many times over the past century, Congress has confronted a new technology and has had to decide whether, and how, to preempt state authority to regulate it.

    • Yiqin Fu
  • Abstract image for artificial intelligence
    research
    AI Agents and Democratic Resilience

    Autonomous AI agents are arriving at a moment of acute vulnerability for liberal democratic orders. Amid deep uncertainty about agents’ prospective capabilities and impacts, this essay considers how they might both accelerate and mitigate the structural pressures loosening democracy’s screws—and how to ensure the robust protection of democratic values in the age of agentic AI. 

    • Seth Lazar
    Knight Foundation
  • People walking on wet road during rainy season in Sweden
    article
    Participatory War and Its Challenges for Total Defense

    As governments promote digital civic engagement in defense, they must acknowledge the possible deterioration of civil liberties and the legal vulnerability of civilians acting as potential combatants.

    • Kristin Ljungkvist