This dataset contains structured annotations for 30 prominent AI agents released or actively developed in 2025, compiled as part of the AI Agent Index project. The AI Agent Index is a systematic effort to catalogue and characterise real-world AI agents across dimensions relevant to accountability, safety, and transparency. File Contents File Description 2025_annotations.json Full annotations in nested JSON format, preserving the hierarchical section structure with inline source links and archived URLs. Text is in Markdown format. 2025_annotations.csv Flattened tabular version with one row per agent and one column per field. Text is in Markdown format. Data Structure Each agent record is organised into seven thematic sections: Inclusion criteria: the signals used to select the agent (search volume, market cap, GitHub stars, developer importance) Product overview: agent name, description, release date, advertised use case, pricing, target users, website, and category Company & accountability: developer identity, legal entity, place of incorporation, profit status, parent company, governance documents, AI safety frameworks, and standards compliance Technical capabilities & system architecture: underlying model, documentation, observation space, action space, memory architecture, user interface design, user roles, and openness of components Autonomy & control: autonomy level (L1–L5 scale), human approval requirements, execution monitoring and traceability, emergency stop mechanisms, and usage statistics Ecosystem interaction: agent self-identification to humans and systems, interoperability standards (MCP, A2A, ACP, AGNTCY), and web conduct Safety, evaluation & impact: technical guardrails, sandboxing approaches, risk evaluations, internal and third-party safety testing, benchmark results, vulnerability disclosure programmes, and known incidents More Information For further details about the AI Agent Index project, methodology, and interactive data explorer, visit https://aiagentindex.mit.edu/