principal-agent-alignment

To prevent AI intent drift during long-running task execution, implement continuous alignment checks that detect divergence between interpreted and intended goals, using a lightweight pre-task alignment contract to establish baseline intent.

Human-AI interaction alignment is the continuous matching of AI outputs with actual human intent. You need to monitor it during multi-step tasks, evolving context projects, and multi-stakeholder coordination where interpretation drift is likely.

Set up a pre-task alignment contract by defining lightweight rules for execution, including one clarifying question for drift signaling and proactive surfacing to realign Claude with true intent before proceeding.

Yes, drift signaling realigns Claude without stopping execution by surfacing a single clarifying question when divergence is detected, allowing continuous alignment checks to correct the task trajectory proactively.

Continuous alignment monitoring is specifically designed for tasks with evolving context, detecting drift between Claude's interpretation and your actual intent throughout execution to ensure outputs remain relevant.

Claude misinterprets task intent when context evolves during execution. Alignment rules fix this by requiring proactive surfacing of drift and a single clarifying question to realign before unsafe or irrelevant actions occur.