Contextual Value Alignment: Mapping Human Values to Adaptive Agentic Behaviors Across Deployment Contexts

TL;DR: How can we make sure AI agents behave according to human values (like safety or reliability) even as their deployment context changes? Let’s create a framework that dynamically maps values to behaviors, so agentic systems “know” how to adapt.

Research Question: How can agentic AI systems dynamically interpret and operationalize core human values (e.g., safety, reliability) in response to changes in deployment context (e.g., CLI vs. embedded gateway)?

Hypothesis: A context-aware value alignment framework will enable AI agents to consistently uphold human values, even as deployment contexts and operational constraints evolve.

Experiment Plan: Analyze value-to-behavior mappings in Claude Code and OpenClaw across multiple deployment scenarios. Develop a flexible mapping engine that links high-level values to low-level design choices (e.g., switching between per-action and perimeter-level safety based on context). Test the system in environments where deployment context shifts (e.g., local development vs. cloud gateway) and measure adherence to value-driven policies, error rates, and user trust.

References:

• Liu, J., Zhao, X., Shang, X., & Shen, Z. (2026). Dive into Claude Code: The Design Space of Today's and Future AI Agent Systems.
• Lee, C. P. (2024). Design, Development, and Deployment of Context-Adaptive AI Systems for Enhanced User Adoption. CHI Extended Abstracts.