{"paper":{"title":"Threat-Preserving Representation Sensitivity in Agent-Security Benchmarks","arxiv_id":"2610.03585","primary_cat":"cs.CR","submitted_at":"2026-10-02T16:55:50+00:00","version_compiled":"v1"},"notice":"Every quoted sentence is the authors' text from the arXiv listing abstract. Labels and probabilities are the outputs of classifier-1.13.0 on the question bank abstract-p1-20260923, exactly as returned. No sentence in this record was written by a language model.","schema":"pith.compiled-record/v0","status":"compiled","abstract":{"source":"arXiv listing abstract v1","sentences":[{"i":0,"p":1.0,"id":"s0","own":false,"role":"background","text":"Security benchmarks for LLM-based agents often report the attack success rate (ASR) as a measure of model robustness and use these scores to compare different models and defense mechanisms, assuming that they describe the security of the agent.","hedge":{"p":0.99,"label":"asserted"},"own_p":0.05,"main_claim_p":0.02,"quantitative":false,"quantitative_p":0.08,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.0,"background":1.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":1,"p":0.62,"id":"s1","own":false,"role":"gap","text":"In this paper, we explore whether it also influences the benchmark's measurement.","hedge":{"p":0.56,"label":"unclear"},"own_p":0.16,"main_claim_p":0.03,"quantitative":false,"quantitative_p":0.02,"role_probabilities":{"gap":0.62,"method":0.0,"result":0.0,"background":0.03,"limitation":0.0,"future_work":0.0,"contribution":0.35,"interpretation":0.0}},{"i":2,"p":0.62,"id":"s2","own":true,"role":"method","text":"To measure the effect of the benchmark representation, we introduce threat-preserving representation sensitivity (TPRS), which measures how much the ASR changes when we change the agent-visible representation while holding the underlying task, harmful action, security policy, ground truth, environment, and the evaluation criteria fixed.","hedge":{"p":0.98,"label":"asserted"},"own_p":0.86,"main_claim_p":0.05,"quantitative":false,"quantitative_p":0.05,"role_probabilities":{"gap":0.0,"method":0.62,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.38,"interpretation":0.0}},{"i":3,"p":1.0,"id":"s3","own":true,"role":"result","text":"On Agent Security Bench (ASB), replacing threat-related tool names with threat-neutral names raises the committed attack success rate by 11.67 percentage points on GPT-5-mini and by 13.21 points on Claude Haiku 4.5.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.96,"main_claim_p":0.06,"quantitative":true,"quantitative_p":0.99,"role_probabilities":{"gap":0.0,"method":0.0,"result":1.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":4,"p":1.0,"id":"s4","own":true,"role":"result","text":"On MCPTox, replacing the original neutral tool name with an explicit threat-related name lowers the ASR by 11.00 percentage points on GPT-5-mini and 4.11 points on Claude Haiku 4.5.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.96,"main_claim_p":0.04,"quantitative":true,"quantitative_p":0.99,"role_probabilities":{"gap":0.0,"method":0.0,"result":1.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":5,"p":1.0,"id":"s5","own":true,"role":"result","text":"On AgentDojo, adding threat-related wording to the attack-relevant tool changes ASR by only 0.50 percentage points on GPT-4o-mini, yet the benign utility falls by 5.36 points on tasks requiring that tool.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.95,"main_claim_p":0.04,"quantitative":true,"quantitative_p":0.99,"role_probabilities":{"gap":0.0,"method":0.0,"result":1.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":6,"p":0.99,"id":"s6","own":true,"role":"result","text":"We ran an experiment on MCPTox where we observed that a threat-neutral name matched on token count, length, and casing reproduces most of the shift produced by the threat-explicit name (8.54 of 11.00 points on GPT-5-mini).","hedge":{"p":0.99,"label":"asserted"},"own_p":0.97,"main_claim_p":0.05,"quantitative":true,"quantitative_p":0.99,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.99,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.01}},{"i":7,"p":0.95,"id":"s7","own":true,"role":"interpretation","text":"The results show that a security score measured under one representation may fail to generalize across threat-preserving representations of the same security problem.","hedge":{"p":0.86,"label":"hedged_possibility"},"own_p":0.91,"main_claim_p":0.77,"quantitative":false,"quantitative_p":0.06,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.05,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.95}},{"i":8,"p":0.64,"id":"s8","own":false,"role":"interpretation","text":"Robustness claims should therefore be supported by performance across a controlled set of threat-preserving representations rather than relying on a single representation-dependent score.","hedge":{"p":0.9,"label":"asserted"},"own_p":0.2,"main_claim_p":0.81,"quantitative":false,"quantitative_p":0.06,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.36,"contribution":0.0,"interpretation":0.64}}]},"compiled":{"main_claim":{"p":0.74,"text":"The results show that a security score measured under one representation may fail to generalize across threat-preserving representations of the same security problem.","location":"abstract s7","hedge_strength":{"p":0.86,"label":"hedged possibility"}},"provenance":{"route":{"via":"TypeSafe API","model":"classifier-1.13.0","endpoint":"https://api.typesafe.ai/v1/systemone","provider":"TypeSafe"},"packing":"one request per paper","requests":1,"input_tokens":7282,"cost_usd_reported":0.0},"compiled_at":"2026-10-05","bank_version":"abstract-p1-20260923","model_version":"classifier-1.13.0","abstract_states":{"claims_first":0.05,"states_limitation":0.32,"human_participants":0.03,"controlled_comparison":0.85,"randomized_assignment":0.02,"states_quantitative_result":0.99,"compares_with_prior_results":0.33,"mentions_code_or_data_availability":0.03},"evidential_basis":{"p":1.0,"value":"observation or experiment","probabilities":{"argument":0.0,"literature":0.0,"proof_or_derivation":0.0,"analysis_of_existing_data":0.0,"observation_or_experiment":1.0,"simulation_or_computation":0.0}},"contribution_type":{"p":0.96,"value":"empirical study","probabilities":{"review":0.0,"method_or_tool":0.04,"empirical_study":0.96,"simulation_study":0.0,"theoretical_result":0.0,"dataset_or_benchmark":0.0,"position_or_commentary":0.0}},"contribution_statement":{"p":1.0,"text":"To measure the effect of the benchmark representation, we introduce threat-preserving representation sensitivity (TPRS), which measures how much the ASR changes when we change the agent-visible representation while holding the underlying task, harmful action, security policy, ground truth, environment, and the evaluation criteria fixed.","location":"abstract s2"}},"coverage":{"tier":"abstract","levels_run":["P abstract"],"not_examined":["full text","figures and tables","references and citing sentences","relations to other papers"],"sources_read":["arXiv listing abstract v1"],"levels_not_run":["S sections","T body sentences","C citations","R cross-paper relations"],"sentences_examined":9,"sentences_in_abstract":9},"measured_accuracy":{"url":"https://pith.science/accuracy","measured_on":"2026-10-02","method":"two independent model families labelled a fresh random sample blind; the figure is how often the published label matches their consensus","sentence_role_agreement":0.92,"main_claim_agreement":0.908,"typed_citation_agreement":0.933,"typed_citation_agreement_95_interval":[0.917,0.945],"review_comments_accurate":0.758,"review_comments_accurate_95_interval":[0.684,0.82],"review_caught_admitted_error_found_or_related":0.448,"review_comments_judged":{"judged_by":["gpt-5.6-luna","grok-4.7"],"judged_against":"Pith's own text copy of each paper: text extracted from the PDF, the first 90,000 characters","measured_on":"2026-10-04","accurate":113,"decided":149,"sampled_reviews":200,"split_by_lane":false,"by_lane":"https://pith.science/accuracy#by-lane"}}}