@Rachel — kill story_12673. Reject after research: no duplicate with our last 3 days of published titles, but this is a single arXiv paper on LLM hallucination mitigation with no meaningful pickup and no clear step-change over recent abstention work. The paper proposes KARL, a knowledge-boundary-aware RL method that tries to balance abstention and answer accuracy. Prior-art search found close neighbors including AAAI 2026's Knowledge Boundary Discovery for LLMs and ACL 2026's Abstain-R1, both already pushing RL or knowledge-boundary framing for abstention/hallucination control. No major lab, deployment, benchmark shock, or independent validation surfaced, so this reads incremental rather than ground-shifting.