{
  "abstract": "Genomic language models generate long nucleotide continuations from short prefixes, exposing a direct route for studying training-sequence extraction through the generation interface. We present an output-only black-box attack on Evo 2 that selects convergent groups of repeated generations without gradients, token likelihoods, or training-corpus access during candidate selection. Recovered content is then scored by an evaluator combining aligned length, nucleotide identity, match multiplicity, and sequence complexity. Across 304 runs on Evo 2 7B with bacteriophage and plasmid prefixes, the attack yields 888 candidate clusters, of which 85 match training segments of at least 500 bp, giving a cluster-level attack success rate of 9.57%. Final filtering retains 22 uniquely matched extraction events at the 500-bp threshold and 15 at 1,000 bp, covering 14 and 7 distinct training-sequence regions, respectively. Within-category source-homology screening retains 11 of the 22 events at a 99% exclusion threshold and four at 98%, with the latter mapping to two reference regions. The results establish convergence among repeated nucleotide generations as a practical output-only signal for recovering long training-sequence segments and characterize how recovery depends on prefix length, cluster size, perplexity, early stopping, and sequence domains.",
  "abstract_shown": true,
  "affiliations": [],
  "article_type": null,
  "arxiv": "2610.15396",
  "author_total": 6,
  "authors": [
    {
      "name": "Tianhao Mao"
    },
    {
      "name": "Zihao Wang"
    },
    {
      "name": "Dongfang Zhao"
    },
    {
      "name": "Rui Zhu"
    },
    {
      "name": "Xiaofeng Wang"
    },
    {
      "name": "Haixu Tang"
    }
  ],
  "bibtex": "https://research.draconitic.com/paper/2f0y4375kbpyw.bib",
  "cited_by": 0,
  "doi": "10.48550/arxiv.2610.15396",
  "fulltext_url": null,
  "id": "2f0y4375kbpyw",
  "issns": [],
  "landing_url": "https://arxiv.org/abs/2610.15396",
  "language": null,
  "licence": "CC-BY-4.0",
  "licence_label": "CC BY 4.0",
  "licence_source": "article",
  "licence_url": "http://creativecommons.org/licenses/by/4.0/",
  "metadata_licence": "CC0-1.0",
  "native_subjects": [
    [
      "arxiv",
      "cs.CR"
    ],
    [
      "arxiv",
      "q-bio.GN"
    ],
    [
      "openalex",
      "1702|Adversarial Robustness in Machine Learning"
    ]
  ],
  "openalex": "W7221790718",
  "pdf_url": "https://arxiv.org/pdf/2610.15396",
  "pmcid": null,
  "pmid": null,
  "published": "2026-10-11",
  "publisher": null,
  "ris": "https://research.draconitic.com/paper/2f0y4375kbpyw.ris",
  "sources": [
    {
      "first_seen": "2026-10-30T12:29:21Z",
      "native_id": "2610.15396",
      "source": "arxiv",
      "url": "https://arxiv.org/abs/2610.15396"
    }
  ],
  "subjects": [
    {
      "field": "biology",
      "origin": "arxiv:q-bio.GN",
      "subfield": "genetics-genomics"
    }
  ],
  "title": "Training-Data Extraction from Genomic Language Models",
  "title_text": "Training-Data Extraction from Genomic Language Models",
  "type": "preprint",
  "updated_at": "2026-10-30T12:29:21Z",
  "url": "https://research.draconitic.com/paper/2f0y4375kbpyw",
  "venue": "arXiv",
  "versions": []
}