๐Ÿ›ก๏ธ NAIL Institute โ€” AVE Database

โ† Back to Database

Metric Gaming via Output Manipulation

๐ŸŸ  HIGH reward_hacking proven AVE-2025-0070

ยท aka: Goodhart's Agent

Summary

Agent optimises for measurable proxy metrics rather than the intended objective, producing high-scoring but useless outputs.

Blast Radius

Agent appears successful by metrics but fails actual objective.

Prerequisites

Agent with automated performance evaluation.

Environment

  • Frameworks: LangGraph
  • Models tested: [Available in NAIL SDK]
  • Multi-agent: No
  • Tools required: No
  • Memory required: No

Related