{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:VW7GUEVKB2B5FHYM3W53XFX7XG","short_pith_number":"pith:VW7GUEVK","schema_version":"1.0","canonical_sha256":"adbe6a12aa0e83d29f0cddbbbb96ffb990d7c7bd2d6d5c49b9745477283e772e","source":{"kind":"arxiv","id":"2605.14240","version":1},"attestation_state":"computed","paper":{"title":"Paraphrasing Attack Resilience of Various AI-Generated Text Detection Methods","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Binoculars-inclusive ensembles detect AI text most accurately but lose the largest share of that accuracy when text is paraphrased.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Andrii Shportko, Inessa Verbitsky","submitted_at":"2026-05-14T01:12:34Z","abstract_excerpt":"The recent large-scale emergence of LLMs has left an open space for dealing with their consequences, such as plagiarism or the spread of false information on the Internet. Coupling this with the rise of AI detector bypassing tools, reliable machine-generated text detection is in increasingly high demand. We investigate the paraphrasing attack resilience of various machine-generated text detection methods, evaluating three approaches: fine-tuned RoBERTa, Binoculars, and text feature analysis, along with their ensembles using Random Forest classifiers. We discovered that Binoculars-inclusive ens"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":true,"formal_links_present":false},"canonical_record":{"source":{"id":"2605.14240","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-14T01:12:34Z","cross_cats_sorted":[],"title_canon_sha256":"b821678c4f2c03efd2dc8f0b723d3c97c9afd611b1132d621868118df7a207cc","abstract_canon_sha256":"7db814a6ef223f5f9f1aca2780bb091586e7f526b363973b21691e363e2f8253"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:39:10.660215Z","signature_b64":"uLLlS65dUba5FQKStGkN6N1MM0S6aJkwtRveowHBCjG1dVkweJdRbhCSsSdr7wcsUnY4Fv60Sh0kA1VZGp8cDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"adbe6a12aa0e83d29f0cddbbbb96ffb990d7c7bd2d6d5c49b9745477283e772e","last_reissued_at":"2026-05-17T23:39:10.659749Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:39:10.659749Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Paraphrasing Attack Resilience of Various AI-Generated Text Detection Methods","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Binoculars-inclusive ensembles detect AI text most accurately but lose the largest share of that accuracy when text is paraphrased.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Andrii Shportko, Inessa Verbitsky","submitted_at":"2026-05-14T01:12:34Z","abstract_excerpt":"The recent large-scale emergence of LLMs has left an open space for dealing with their consequences, such as plagiarism or the spread of false information on the Internet. Coupling this with the rise of AI detector bypassing tools, reliable machine-generated text detection is in increasingly high demand. We investigate the paraphrasing attack resilience of various machine-generated text detection methods, evaluating three approaches: fine-tuned RoBERTa, Binoculars, and text feature analysis, along with their ensembles using Random Forest classifiers. We discovered that Binoculars-inclusive ens"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"We discovered that Binoculars-inclusive ensembles yield the strongest results, but they also suffer the most significant losses during attacks.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the paraphrasing attacks and evaluation datasets used are representative of real-world evasion attempts and that the reported performance differences generalize beyond the specific test conditions.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Binoculars-inclusive ensembles detect AI text best overall but suffer the largest performance drops under paraphrasing attacks.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Binoculars-inclusive ensembles detect AI text most accurately but lose the largest share of that accuracy when text is paraphrased.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"6d6f3fd0bfdb4a45d34b1b103b6c6dcfdb0b90da5892a46a400440f0f7b91d41"},"source":{"id":"2605.14240","kind":"arxiv","version":1},"verdict":{"id":"a2d7c01c-0cce-4d2a-9031-e0414f2f6611","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T01:31:41.008483Z","strongest_claim":"We discovered that Binoculars-inclusive ensembles yield the strongest results, but they also suffer the most significant losses during attacks.","one_line_summary":"Binoculars-inclusive ensembles detect AI text best overall but suffer the largest performance drops under paraphrasing attacks.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the paraphrasing attacks and evaluation datasets used are representative of real-world evasion attempts and that the reported performance differences generalize beyond the specific test conditions.","pith_extraction_headline":"Binoculars-inclusive ensembles detect AI text most accurately but lose the largest share of that accuracy when text is paraphrased."},"references":{"count":14,"sample":[{"doi":"","year":2025,"title":"https:// www.gptinf.com/","work_id":"2e0dbba6-3461-4998-a815-71ecd59fb670","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"arXiv preprint arXiv:2309.07755 (2023)","work_id":"aeb1d0a1-4431-4f13-a72f-b5b25616cd58","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"On the possibilities of ai-generated text detection","work_id":"ecd9bcaf-8465-4a71-8229-27cc0d08920d","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Epub 2023 Jun","work_id":"a1073dbc-3abe-4d4b-a3df-d721ea4eef47","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Xiaomeng Hu, Pin-Yu Chen, and Tsung-Yi Ho","work_id":"4e95670f-355f-4cfb-afa8-f1a708ad2457","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":14,"snapshot_sha256":"7e838583f3605f12f7208c82a4a4a9b84fb3ed3d002b288ee7662ee3f1cd52f9","internal_anchors":1},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2605.14240","created_at":"2026-05-17T23:39:10.659820+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.14240v1","created_at":"2026-05-17T23:39:10.659820+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.14240","created_at":"2026-05-17T23:39:10.659820+00:00"},{"alias_kind":"pith_short_12","alias_value":"VW7GUEVKB2B5","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_16","alias_value":"VW7GUEVKB2B5FHYM","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_8","alias_value":"VW7GUEVK","created_at":"2026-05-18T12:33:37.589309+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/VW7GUEVKB2B5FHYM3W53XFX7XG","json":"https://pith.science/pith/VW7GUEVKB2B5FHYM3W53XFX7XG.json","graph_json":"https://pith.science/api/pith-number/VW7GUEVKB2B5FHYM3W53XFX7XG/graph.json","events_json":"https://pith.science/api/pith-number/VW7GUEVKB2B5FHYM3W53XFX7XG/events.json","paper":"https://pith.science/paper/VW7GUEVK"},"agent_actions":{"view_html":"https://pith.science/pith/VW7GUEVKB2B5FHYM3W53XFX7XG","download_json":"https://pith.science/pith/VW7GUEVKB2B5FHYM3W53XFX7XG.json","view_paper":"https://pith.science/paper/VW7GUEVK","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.14240&json=true","fetch_graph":"https://pith.science/api/pith-number/VW7GUEVKB2B5FHYM3W53XFX7XG/graph.json","fetch_events":"https://pith.science/api/pith-number/VW7GUEVKB2B5FHYM3W53XFX7XG/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/VW7GUEVKB2B5FHYM3W53XFX7XG/action/timestamp_anchor","attest_storage":"https://pith.science/pith/VW7GUEVKB2B5FHYM3W53XFX7XG/action/storage_attestation","attest_author":"https://pith.science/pith/VW7GUEVKB2B5FHYM3W53XFX7XG/action/author_attestation","sign_citation":"https://pith.science/pith/VW7GUEVKB2B5FHYM3W53XFX7XG/action/citation_signature","submit_replication":"https://pith.science/pith/VW7GUEVKB2B5FHYM3W53XFX7XG/action/replication_record"}},"created_at":"2026-05-17T23:39:10.659820+00:00","updated_at":"2026-05-17T23:39:10.659820+00:00"}