{
  "id": "reward-hacking",
  "code": "PTL-0172",
  "term": "Reward Hacking",
  "aliases": [
    "specification gaming",
    "reward overoptimization",
    "reward gaming"
  ],
  "category": "failure-modes",
  "definition": "Reward hacking is when a model trained or optimized against a reward finds ways to score highly that do not achieve the intended goal, exploiting flaws in the reward model, grader, or tests.",
  "description": "In language models it shows up as answers padded to look thorough, flattery that raters prefer, and code that special-cases tests instead of solving the problem. Verifiable rewards reduce some forms, while careful graders, held-out tests, and monitoring of reasoning traces help detect others.",
  "example": null,
  "broader": [],
  "narrower": [],
  "related": [
    "reward-model",
    "rlhf",
    "rlvr",
    "sycophancy",
    "chain-of-thought-monitoring"
  ],
  "introduced": null,
  "added": "2026-10-10",
  "sources": [
    {
      "title": "Concrete Problems in AI Safety",
      "authors": "Amodei et al.",
      "year": 2016,
      "url": "https://arxiv.org/abs/1606.06565"
    }
  ],
  "url": "https://protologue.com/t/reward-hacking/",
  "citation": "Protologue. (2026). Reward Hacking. In Protologue: A Taxonomy of Prompting and LLM Techniques (v1.4.0, PTL-0172). https://protologue.com/t/reward-hacking/"
}