{"paper":{"title":"Inducing Overthink: Hierarchical Genetic Algorithm-based DoS Attack on Black-Box Large Language Reasoning Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"A hierarchical genetic algorithm can force large reasoning models to generate up to 26 times longer outputs by perturbing input logic.","cross_cats":["cs.AI"],"primary_cat":"cs.CR","authors_text":"Hui Xue, Jialing Tao, Jiaqi Weng, Licheng Pan, Shuqiang Wang, Wei Cao, Zhixuan Chu","submitted_at":"2026-05-13T10:57:10Z","abstract_excerpt":"Large Reasoning Models (LRMs) are increasingly integrated into systems requiring reliable multi-step inference, yet this growing dependence exposes new vulnerabilities related to computational availability. In particular, LRMs exhibit a tendency to \"overthink\", producing excessively long and redundant reasoning traces, when confronted with incomplete or logically inconsistent inputs. This behavior significantly increases inference latency and energy consumption, forming a potential vector for denial-of-service (DoS) style resource exhaustion. In this work, we investigate this attack surface an"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Across four state-of-the-art reasoning models, the proposed method substantially amplifies output length, achieving up to a 26.1x increase on the MATH benchmark and consistently outperforming benign and manually crafted missing-premise baselines.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the composite fitness function reliably captures genuine overthinking rather than simply maximizing length through superficial perturbations, and that this behavior generalizes beyond the tested benchmarks and models.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"A hierarchical genetic algorithm induces overthinking in black-box large reasoning models by perturbing logical structure, achieving up to 26.1x longer outputs on the MATH benchmark.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A hierarchical genetic algorithm can force large reasoning models to generate up to 26 times longer outputs by perturbing input logic.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"2aafdbcdc9dc31cea968d874e601f19873f40e00f0a7f61350992771e304243f"},"source":{"id":"2605.13338","kind":"arxiv","version":2},"verdict":{"id":"85dc1a34-5a3d-403d-8f1a-99f57e30efe3","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T05:50:17.728750Z","strongest_claim":"Across four state-of-the-art reasoning models, the proposed method substantially amplifies output length, achieving up to a 26.1x increase on the MATH benchmark and consistently outperforming benign and manually crafted missing-premise baselines.","one_line_summary":"A hierarchical genetic algorithm induces overthinking in black-box large reasoning models by perturbing logical structure, achieving up to 26.1x longer outputs on the MATH benchmark.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the composite fitness function reliably captures genuine overthinking rather than simply maximizing length through superficial perturbations, and that this behavior generalizes beyond the tested benchmarks and models.","pith_extraction_headline":"A hierarchical genetic algorithm can force large reasoning models to generate up to 26 times longer outputs by perturbing input logic."},"references":{"count":37,"sample":[{"doi":"","year":2021,"title":"2021 IEEE Symposium on Security and Privacy (SP) , pages =","work_id":"c7d21920-cd15-4565-98ff-3e7e963e435b","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"arXiv e-prints , year =","work_id":"9316a530-9491-487b-aea2-2ca73c25d175","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Do NOT Think That Much for 2+ 3=? On the Overthinking of o1-Like LLMs , author=. CoRR , year=","work_id":"e21a0546-f840-44bf-b2f4-f35627d858ee","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Missing premise exacerbates overthink- ing: Are reasoning models losing critical thinking skill? arXiv preprint arXiv:2504.06514","work_id":"1529b935-8663-45f4-847e-3c0ed9965ba2","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"The danger of overthinking: Examining the reasoning-action dilemma in agentic tasks","work_id":"4600dc6b-7c39-4ddb-9872-e03cb8a7efae","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":37,"snapshot_sha256":"fd297f69d655ecc146e9ec17bed26c5bb27b9314492ce47e959c764747d835d5","internal_anchors":7},"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"}