Agentic Trust & Safety

Research on trustworthy and safe AI agents

Research concepts

AI Trust refers here to the reliability, transparency, and verifiability of behavior and outcomes. AI Safety focuses on preventing, detecting, and mitigating harmful behavior and risks. In this research framework, safety is one dimension of AI trust.

Conceptual relationship between AI trust and AI safety In this site’s AI trust framework, AI safety is one dimension alongside reliability, transparency, and verifiability. AI Trust Reliable · transparent · verifiable AI safety Prevent · detect · mitigate
AI safety ⊂ AI Trust

Research framework

A single agent’s decision loop and the communication lifecycle of multiple agents.

01 Single agent

Single-agent planning, action, and feedback loop Planning leads to action; action produces feedback; feedback informs the next plan. Planning Decide Action Execute Feedback Evaluate Self-audit · evidence · defense

02 Multi-agent

Multi-agent communication before, during, and after exchange Before communication, verify identities and permissions; during exchange, monitor messages and coordination; afterward, validate outcomes and trace issues. Before Prepare During Exchange After Review Risk prediction Risk monitoring Risk attribution

Publications

Single-agent research