Metraipolis › archive › dc_381fd329aa26
Reinforcement Learning on Benign Facts Amplifies Leakage of Memorized Private Data — arXiv:2608.21727
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2608.21727
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-18T07:42:23.422Z
- SHA-256 of the archived text
57dfaf4cd91403acdb7484ffc4ce457bddd7b6a3a45a2ec6b919bbd14ab65d49- Stable URL of this page
https://metraipolis.com/documento/dc_381fd329aa26
How to cite
Reinforcement Learning on Benign Facts Amplifies Leakage of Memorized Private Data — arXiv:2608.21727 — arXiv (en), https://arxiv.org/abs/2608.21727 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_381fd329aa26 (Metraipolis archive dc_381fd329aa26, sha256 57dfaf4cd91403acdb7484ffc4ce457bddd7b6a3a45a2ec6b919bbd14ab65d49)
BibTeX
@misc{dc381fd329aa26,
title = {Reinforcement Learning on Benign Facts Amplifies Leakage of Memorized Private Data — arXiv:2608.21727},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2608.21727},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_381fd329aa26; sha256 57dfaf4cd91403acdb7484ffc4ce457bddd7b6a3a45a2ec6b919bbd14ab65d49},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.21727. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Reinforcement learning with verifiable rewards (RLVR) is deployed to make models better at reasoning tasks, but its side effect on what models will divulge is under studied. Here we show that RLVR on facts increases extraction of personally identifiable information (PII) the instruct model had already memorized. We first confirm that instruct models have already memorized PII but leave them latent, rarely surfacing one when asked. We then apply RL on benign factual data that contains no PII of any kind, and re-probe: a targeted probe over name->email pairs, and an untargeted free-recall prompt that simply asks the model to list the addresses it knows. PII extraction rises sharply under both: on DeepSeek-V3.1, verbatim recall@k increases from 0.155 to 0.370, a 2.4x gain. The effect scales with model size: across three models spanning 8B to 671B parameters, absolute leakage is largest in the biggest model. Meanwhile model's reasoning abilities and refusal rates are retained, indicating that RL selectively changes which memorized information is accessible rather than broadly altering the model. In summary, memorized private data can be made markedly more extractable by training that never touches it. This gives an adversary a route to memorized data that requires no privacy-relevant training signal and no access to the data itself -- only the ability to fine-tune on something innocuous. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-08-25 in cs.AI). Categorie arXiv: Machine Learning (cs.LG) ; Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2608.21727 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito). Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina. Letto il 2026-10-18T07:37:52.826Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-08-25?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).