Reckoning with AI agents

Report

This report provides a narrative summary of a symposium on AI agents held at Balliol College in the summer of 2025. Operated under the Chatham House Rule, the discussion explored the transition from generative AI to "agentic" AI—systems capable of autonomous action and complex task execution. The report defines the spectrum of agent autonomy and examines the tension between the utility of these systems and the incremental and systemic risks they introduce, particularly regarding transparency, bias, and supply chain concentration. Key themes include the current landscape of hype versus technical reality, the challenges of regulating rapidly evolving technologies (contrasting sectoral vs. blanket approaches), and the integration of agents into high-stakes sectors like finance and healthcare. Furthermore, the report analyses the implications for democratic life and the media ecosystem, highlighting threats to journalism’s financial viability and the potential for exacerbated polarisation or disinformation. Conversely, the summary notes the potential for AI agents to strengthen democratic institutions through deliberation support and fact-checking.

Authors: Simon, FM; Ross Arguedas, A; Eggert, L

• Publication status: Published
• Publisher: Balliol College
• Place of publication: Oxford, UK
• Publication date: 2025-12-17
• DOI: 10.5287/ora-ppgrbwpva
• Language: English
• Keywords: artificial intelligence; AI agents; AI and democracy; AI and work; generative AI; AI; agentic AI