@misc{indiciae89dcd81c0bf4,
  title = {An Empirical Study of Reward Specification and Benchmark Reliability in GRPO-based LLM Unlearning},
  author = {Rubén Balbastre and Juan Manuel Orduña and Mariano Pérez},
  year = {2026},
  url = {https://arxiv.org/abs/2608.17804},
  note = {Source identifier: 2608.17804}
}