{"id":"bc7f140c-cc8b-430f-9eba-588b5da04df4","arxiv_id":"2610.03585","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"Attack-success rates in three LLM-agent security benchmarks shift by up to 13 percentage points when only the threat-related wording the agent sees is changed, showing that benchmark security scores are representation-dependent.","lead":"Security scores for AI agents shift by up to 13 percentage points when only the wording of tool names or descriptions changes, while the underlying threat stays identical. The study warns that a single benchmark number can overstate how robust an agent really is, and that robustness claims need testing across several threat-preserving representations.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek/deepseek-v4-flash","headline":"Temporal/API drift is not controlled for the ASB and MCPTox arm comparisons; the paper's own AgentDojo data show a 5.24-point shift on untouched tasks, so the headline effect sizes need interleaved-run confirmation.","rationale":"The central claim only needs one clean demonstration that ASR moves under a threat-preserving representation change, and the paper's strongest evidence is the ASB A1 and MCPTox N1 contrasts with paired, cluster-bootstrapped confidence intervals excluding zero, robustness to execution-failure treatments, null orthographic controls, and the S3 matched-neutral control. That is genuine independent support. The causal condition that must hold is that each arm comparison isolates the representation change from other time-varying factors. The paper interleaves or randomizes arm order only for AgentDojo; for ASB and MCPTox it does not report run order or timestamps. The AgentDojo +5.24-point utility shift on unchanged tasks demonstrates that API-side drift of several points can occur during this kind of study. If ASB A0 and A1, or MCPTox N0 and N1, ran as separate blocks, the headline effect sizes could be inflated by drift. The reader's semantic-equivalence worry is acknowledged by the authors and mainly affects attribution of the effect to threat wording versus apparent function, not whether the change qualifies as threat-preserving under the paper's definition. The MCPTox subset representativeness limits generalization of the effect size but not the existence of a large effect in those cases. The most decisive unresolved risk is therefore time confounding, and a block-randomized interleaved replication or a timestamp audit would settle it. The reader's CONDITIONAL verdict already captures the need for additional verification, so my read does not change that verdict.","tokens_in":20696,"tokens_out":18324,"duration_ms":231109,"concrete_test":"Recover the per-arm execution order and timestamps for the ASB A0/A1/NULL2 and MCPTox N0/N1/S3 comparisons. If the arms were block-sequential rather than interleaved, re-run the four headline contrasts in short alternating blocks (e.g., A0 and A1 per replicate on the same API version, recording a model-version fingerprint) and recompute the paired differences. If the re-run estimates move toward zero by more than the reported 95% CIs, or change sign, temporal/API drift is a confound and the central claim must be re-scoped; if the effects reproduce within roughly 3 points, the concern is settled.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most load-bearing unresolved risk is temporal/API drift, not semantic equivalence. The paper reports randomizing arm execution order only for AgentDojo (Section III.C) and gives no timestamps or interleaving information for the ASB and MCPTox comparisons. AgentDojo's own data show a +5.24-point benign-utility shift on tasks whose representation was not changed at all, which the authors attribute to \"temporal drift in the model API\" and therefore treat as suggestive rather than established (Section IV.C.b, b303). If ASB A0/A1 or MCPTox N0/N1 were run as blocks at different times, an API-side behavior change could masquerade as representation sensitivity in either direction. The ASB A1 effect (+11.67/+13.21 points) and MCPTox N1 effect (-11.00/-4.11 points) are large enough that a drift of the magnitude already observed in AgentDojo would be a meaningful confound. The orthographic controls and execution-failure robustness checks do not rule this out, because they do not control the time dimension. The semantic-equivalence limitation is real, but it is partly definitional under the paper's threat-preserving definition, which explicitly allows wording meaning to change (Section III.A), and it mostly affects mechanism attribution rather than whether a causal effect was measured. The time confound, if present, threatens the measurement itself.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek/deepseek-v4-flash","summary":"The paper introduces threat-preserving representation sensitivity (TPRS), defined as Δ_T(B,m) = ASR(T(B),m) − ASR(B,m), where T changes only the agent-visible representation while the task, harmful action, security policy, ground truth, environment, and evaluation criteria stay fixed, and measures it on three published agent-security benchmarks across 28,904 runs. On Agent Security Bench (ASB), neutralizing the names of 32 threat-related attacker tools raises committed ASR by 11.67 percentage points on GPT-5-mini (95% CI [4.44, 19.07]) and by 13.21 points on Claude Haiku 4.5 ([6.33, 21.10]), with the effect stable under three treatments of execution failures and under leave-one-tool-out analysis. On MCPTox, replacing neutral tool names with threat-explicit names lowers ASR by 11.00 points on GPT-5-mini ([−15.60, −6.57]) and by 4.11 points on Haiku 4.5, with a form-matched threat-neutral control reproducing 8.54 of the 11.00-point GPT-5-mini shift. On AgentDojo, adding threat wording to the attack-required tool changes ASR by only −0.50 points ([−3.65, +2.51]) while benign utility falls by 5.36 points on tasks using that tool. The methodology includes deterministic substitutions with fixed vocabularies, SHA-256 verification of byte-identical scoring text, cluster-bootstrap confidence intervals at the transformation-application level, Holm correction within the MCPTox hypothesis family, and explicit disclosure of the adaptively added S3 arm.","tokens_in":20996,"tokens_out":18048,"duration_ms":198872,"significance":"This is a useful and well-scoped measurement-validity result for agent-security benchmarking. It is differentiated from prior work on tool-metadata sensitivity (§II.B): the paper holds the security problem, threat, and evaluator fixed and measures instability of the reported score itself, on three independently built benchmarks and two model families, and it shows the effect is benchmark- and metric-dependent rather than a uniform bias. The experimental hygiene is a real strength: byte-identical scoring text verified by SHA-256, deterministic transformation generation with fixed vocabularies, cluster bootstrapping at the unit where the transformation is applied, three treatments of execution failures, leave-one-tool-out robustness, Holm correction for the MCPTox family, and candid disclosure of the adaptively added S3 arm and of the non-representative 203-case MCPTox subset (b203). The S3 matched-neutral control makes the \"threat words alone\" explanation testable, and the finding that 8.54 of 11.00 points of the GPT-5-mini MCPTox shift survives form matching is a falsifiable, reproducible anchor for follow-up work.","major_comments":[{"comment":"The ASB and MCPTox comparisons do not control for temporal or API drift, and the paper's own AgentDojo data show that drift of the relevant magnitude can occur in this setup. Section III.C randomizes arm-execution order only for AgentDojo; for ASB and MCPTox the manuscript reports no timestamps, no batch identifiers, and no interleaving information. Table VIII and Section IV.C.b (sentence b303) report a +5.24-point rise in benign utility on AgentDojo tasks whose representation was not changed, attributed to \"temporal drift in the model API\" and treated as suggestive. Since the ASB A1 lower 95% CI bounds are 4.44 points on GPT-5-mini and 6.33 points on Haiku 4.5 (Table V), a between-block API shift of the magnitude the paper already observed could materially move the measured differences in either direction, and could fully account for the smaller MCPTox Haiku 4.5 effect (4.11 points; CI [−7.55, −0.99], Table VII). The orthographic controls (NULL2, O1, O2) and the execution-failure treatments in Table VI are within-arm robustness checks and do not address the time dimension. I request either a documented execution schedule demonstrating that arms were interleaved, or an interleaved replication of the A0/A1 and N0/N1 contrasts.","section":"III.C, IV.C.b, Table VIII","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The causal status of the ASB effect is not in question, but its interpretation as a threat-preserving effect in the usual reading is weaker than the framing in the abstract and in Section IV.D suggests, and the authors' own acknowledgment of this is confined to Section VI.a. The A1 names were not independently validated for semantic equivalence (b344–b345), so a rename such as DNSPoisoning to DNSConfiguration may change the apparent functionality of the capability, not only its threat signaling; the paper's own attempt at a semantically validated control failed (Section IV.A.d, 10 of 30 names passing the blind check). This is legitimate under the paper's explicit definition, which allows wording meaning to change (b116, Section III.A), and it does not undermine the central measurement claim, but Section IV.D's directional-alignment statement (b49: \"more threat-explicit naming is associated with lower measured ASR in both cases\") overstates what the data establish: on GPT-5-mini, the MCPTox shift is mostly reproduced by the form-matched neutral S3 arm (8.54 of 11.00 points, Table VII), so the alignment with ASB is partly a property of identifier form rather than of threat wording on one of the two models. I recommend that the abstract and Section IV.D carry the Section VI.a caveat explicitly, or that the wording separate the measurement claim (ASR moves under threat-preserving re-representation) from any mechanism claim.","section":"III.A, IV.D, VI.a (b344–b345)","authors_acknowledge":"Thus, transformations may therefore not only alter the appearance of the threat but also the apparent func- tionality of the capability.","authors_acknowledge_p":0.9}],"minor_comments":[{"comment":"The abstract reports the MCPTox headline numbers (11.00 and 4.11 points) without the scope note, disclosed later in Section III.E (b203), that they come from a 203-of-1,348-case subset that is \"not uniformly representative of all possible MCPTox categories\"; the abstract should carry this qualifier.","section":"Abstract and III.E","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"Section III.B.a notes in passing that neutralization in the A1b and A3 arms also rewrote the injected instruction for 4 of the 36 tools (e.g., removing \"discreetly\", replacing \"bypass\" with \"route around\"); since the injected instruction is part of the attack as the agent receives it, these arms go beyond the stated names-or-descriptions scope, and Table II should flag them accordingly, even though the headline A1 arm appears to be clean.","section":"III.B.a","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"ASB and AgentDojo report multiple arm-versus-baseline contrasts without multiplicity correction, yielding 16 ASB contrasts across arms, metrics, and models; a borderline interval such as the Haiku 4.5 A1b native-ASR effect (3.28 points, [0.14, 8.59], Table V) should therefore be described as exploratory, and a sentence stating this would prevent readers from treating every resolved interval as a confirmed finding.","section":"III.F.5, Table V","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"Section III.E states that the MCPTox scorer was validated against published labels with 95.3% recall and a 0.7% false-positive rate, but it does not say whether these statistics were computed on the 203-case subset or on a larger validation set; please state this explicitly.","section":"III.E","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"Figure 2 places all three benchmarks on a common percentage-point axis even though the paper correctly states (Section IV.D) that benchmarks are analyzed independently and are not compared; separate panels with per-benchmark axes would avoid inviting unintended cross-benchmark effect-size comparisons.","section":"Fig. 2","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The extracted text contains missing spaces between words in several places (e.g., \"by11.67percentage\", \"DNSPoisoningbecomes\", \"DNSConfigurationmakes\"), presumably from PDF production; the camera-ready version should fix these rendering artifacts.","section":"Throughout","authors_acknowledge":null,"authors_acknowledge_p":null}],"recommendation":"major_revision"},"author_rebuttal":null,"desk_editor":{"model":"deepseek/deepseek-v4-flash","letter":"The thing to know: this paper actually demonstrates, with intervals that exclude zero, that an agent-security benchmark's attack success rate moves when the same threat is re-represented. Neutralizing threat tool names on ASB raises committed ASR by 11.7–13.2 points on two models; adding threat-explicit names on MCPTox lowers it by 11.0 (GPT-5-mini) and 4.1 (Haiku). The sharpest piece is the S3 control: a threat-neutral name matched on token count, length, and casing reproduces 8.5 of the 11.0 points, so explicit threat vocabulary is not carrying most of the effect. That is a genuinely new empirical result, distinct from the tool-metadata and evaluation-framing work they cite.\n\nThe hygiene is genuinely good. Cluster-bootstrapped paired intervals, SHA-256 checks on scoring text, three treatments for execution failures, leave-one-tool-out robustness, Holm correction on MCPTox, and — most refreshing — open disclosure of the post-hoc S3 arm and the expanded candidate pool. They also correctly keep the AgentDojo utility shift labeled suggestive rather than established. This is how a measurement audit should be written.\n\nNow the soft spots, in proportion. The stress-test concern lands. AgentDojo shows a +5.24-point benign-utility rise on tasks whose representation was not touched, which the authors attribute to API drift. If ASB and MCPTox arms ran as separate blocks at different times, a drift of that magnitude could masquerade as representation sensitivity — the MCPTox Haiku effect (−4.11) is barely larger than the drift they already observed. Their byte-identical replicates (O1/NULL2) bound run-to-run variability at roughly 2 points on GPT-5-mini, which is smaller than the headline effects and counts in the paper's favor, but those replicates do not necessarily bracket the time span between arms. Missing timestamps or interleaving info for ASB/MCPTox is the one fix I would demand.\n\nThe ASB semantic-equivalence limit (DNSPoisoning → DNSConfiguration may alter apparent tool function) is real, but the authors flag it themselves (b344–b345), and it mostly affects mechanism attribution, not whether an effect was measured. The MCPTox 203-case subset is disclosed as non-representative (b203). Those are acknowledged limitations, not hidden flaws.\n\nThis deserves a serious referee. It is a solid, honest contribution that reframes how single-representation ASR scores should be read. The revision needs interleaved-run confirmation or timestamps for the headline arms, plus release of the transformed instances, scorer, and analysis code so independent groups can confirm. If they do that, it is publishable.","headline":"An unusually honest empirical audit showing agent-security ASR is representation-dependent; the one load-bearing fix is interleaved-run or timestamped confirmation against API drift.","tokens_in":21451,"tokens_out":3026,"duration_ms":22356},"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek/deepseek-v4-flash","headline":"Renaming a tool can move an agent-security score by more than 10 points.","keywords":["LLM agents","attack success rate","benchmark validity","representation sensitivity","threat-preserving transformations","tool naming","security evaluation"],"falsifier":"Run the same benchmark arms with a semantically-validated control set of neutral names chosen before any model runs: if a panel of independent raters consistently judges the neutral names as still describing the same tool function, yet the ASR shift disappears, the central claim would weaken from threat-preserving representation sensitivity to changes in apparent tool functionality. Conversely, re-running the MCPTox matched-neutral arm on the full 1,348-case scored subset would test whether the 11-point effect survives the non-representative 203-case subset.","tokens_in":22074,"feed_emoji":"🎯","tokens_out":4967,"duration_ms":41004,"temperature":0.7,"pith_summary":"This paper asks whether a security benchmark for LLM agents measures the model's robustness, or the benchmark's own representation of the threat. The authors introduce threat-preserving representation sensitivity (TPRS), defined as the change in attack success rate when the agent-visible wording changes while the underlying task, harmful action, security policy, ground truth, environment, and evaluation criteria stay fixed. Across 28,904 agent runs on three benchmarks, they find that merely renaming tools moves ASR substantially and in both directions: removing threat-related wording raises ASR on Agent Security Bench by 11.67 percentage points on GPT-5-mini and 13.21 on Claude Haiku 4.5, while adding threat-related wording lowers ASR on MCPTox by 11.00 and 4.11 points. The consequence the paper argues for is that a single-representation security score may not generalize to other threat-preserving representations of the same security problem, so robustness claims should be supported by a controlled set of representations rather than one score.","feed_headline":"Renaming a tool can move a security score by 13 points","feed_subtitle":"Merely rewording tool names changes agent attack success rates; one benchmark score may not generalize.","key_machinery":"The central object is threat-preserving representation sensitivity (TPRS), a measurement defined as $\\Delta_T(B,m) = \\mathrm{ASR}(T(B),m) - \\mathrm{ASR}(B,m)$, where $B$ is a benchmark, $m$ is a model, ASR is the attack success rate, and $T$ is a transformation that changes only the agent-visible representation while keeping the environment, attacker objective, security policy, success criterion, and evaluation mechanism fixed. The argument is carried by systematic, pre-specified renames of tool names and descriptions on three benchmarks with different attack architectures, plus control arms that vary identifier form without adding threat words.","core_discovery":"The central claim is that a security score measured under one representation may fail to generalize across threat-preserving representations of the same security problem. The paper establishes this by defining TPRS as $\\Delta_T(B,m) = \\mathrm{ASR}(T(B),m) - \\mathrm{ASR}(B,m)$, where $T$ changes only the agent-visible representation, and then measuring it on three published benchmarks. On ASB, changing threat-related attacker tool names to neutral names increased committed ASR by 11.67 points on GPT-5-mini and 13.21 on Claude Haiku 4.5; on MCPTox, adding explicit threat wording to neutral names lowered ASR by 11.00 points on GPT-5-mini and 4.11 on Haiku 4.5; on AgentDojo, the ASR change was small at 0.50 points, yet benign utility on tasks requiring the renamed tool fell by 5.36 points. The paper further shows that threat vocabulary alone does not explain the MCPTox effect: a matched threat-neutral name reproducing most of the shift on GPT-5-mini, 8.54 of the 11.00 points.","pith_inferences":["Trainers and deployers could use TPRS-style variation as a cheap robustness screen: rerun a benchmark on a small family of renamed tool identifiers and treat large score movement as evidence that the reported score is partly a representation artifact.","The results suggest an attack-side interpretation: when an attacker controls tool naming, as in ASB and MCPTox, choosing a neutral-sounding name becomes part of the attack, and the gap of 11 to 13 points may understate the ASR an attacker can achieve for free.","A testable extension is checking whether model rankings reverse: with enough models with headroom, rename-based sensitivity might change the ordering implied by a single benchmark score, which the paper leaves as future work.","The matched-neutral control in MCPTox suggests that identifier length, token count, and casing can carry much of the effect; future work could isolate which of these properties drives agent behavior."],"forward_implications":["If the paper is right, published ASR comparisons between models or defenses hold only for the particular representation those benchmarks happened to choose, and could change under another representation.","Robustness claims should report performance across a controlled set of threat-preserving representations, not a single score.","Since representation changes can amplify ASR on ASB or suppress it on MCPTox, benchmark designers should treat representation choices as part of the evaluation design and disclose their sensitivity.","Because native and committed ASR respond differently in ASB, the definition of attack success is itself part of the measurement and should be reported explicitly.","Benign utility may carry representation effects even when ASR looks stable, as in AgentDojo, so security-only evaluations can miss the impact of interface choices."],"supporting_citations":[{"why":"ASB is the first benchmark used; its attacker tool names with threat-related wording are the target of the cue-altering transformation.","marker":"[1]"},{"why":"MCPTox is the second benchmark used; its neutral tool names are replaced with threat-explicit and matched-neutral names.","marker":"[2]"},{"why":"AgentDojo is the third benchmark used; its benign tool names are renamed to test ASR and benign utility effects.","marker":"[3]"},{"why":"Shows that modifying tool descriptions without changing functionality can change tool usage, motivating the claim that metadata choices can affect agent behavior.","marker":"[4]"},{"why":"Shows that tool selection depends on how well a query matches exposed tool metadata, supporting the premise that representation affects agent actions.","marker":"[5]"},{"why":"Establishes a direct link between representation and safety by showing that schema-formatted tool specifications can weaken refusals and increase unsafe tool execution.","marker":"[6]"},{"why":"The Agentic Benchmark Checklist provides the broader benchmark-validity framing that motivates auditing an agent-security benchmark as a measurement instrument.","marker":"[18]"}],"fun_headline_variants":["Tool renaming shifts agent safety scores by 13 points","Same threat, new name: ASR jumps 11–13 points","Threat-preserving renames skew agent security results","Security scores unstable under tool-name changes"],"cache_read_input_tokens":0,"weakest_assumption_plain":"The claimed tool renames are genuinely threat-preserving, meaning the new name changes only what the agent reads, not the apparent function or detectability of the tool itself; the authors acknowledge in Section VI.a that the ASB A1 names were not validated for semantic equivalence and may also alter the apparent functionality of the capability, and the MCPTox results rest on a 203-of-1348-case subset the authors call not uniformly representative.","fun_headline_variants_meta":{"raw":{"variants":["Tool renaming shifts agent safety scores by 13 points","Same threat, new name: ASR jumps 11–13 points","Threat-preserving renames skew agent security results","Security scores unstable under tool-name changes"]},"model":"deepseek/deepseek-v4-flash","effort":"low","cost_usd":0.000171,"raw_usage":{"cost":0.000171248,"is_byok":false,"cost_details":{"upstream_inference_cost":0.000171248,"upstream_inference_prompt_cost":0.000132328,"upstream_inference_completions_cost":3.892e-05},"total_tokens":1289,"prompt_tokens":1150,"completion_tokens":139,"prompt_tokens_details":{"audio_tokens":0,"video_tokens":0,"cached_tokens":256,"cache_write_tokens":0},"completion_tokens_details":{"audio_tokens":0,"image_tokens":0,"reasoning_tokens":73}},"tokens_in":1150,"tokens_out":139,"cost_source":"openrouter_usage","duration_ms":2257,"image_pages":[],"temperature":1.0,"billed_attempts":1,"reasoning_tokens":73,"transport_provider":"openrouter","cache_read_input_tokens":0,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-10-05T06:02:52.087699+00:00","model_set":{"reader":"deepseek/deepseek-v4-flash"},"pipeline_version":"pith-pipeline@v1.4.1-daily-deepseek-or-medium","input_kind":null,"input_chars":null,"input_total_chars":null,"passes_missing":null,"published_standing":null,"published_confidence":null,"objection_audit":null,"run_record":null,"basis":{"kind":"not_recorded","source":null,"chars_given":null,"chars_total":null},"receipt":null,"falsifier":"Run the same benchmark arms with a semantically-validated control set of neutral names chosen before any model runs: if a panel of independent raters consistently judges the neutral names as still describing the same tool function, yet the ASR shift disappears, the central claim would weaken from threat-preserving representation sensitivity to changes in apparent tool functionality. Conversely, re-running the MCPTox matched-neutral arm on the full 1,348-case scored subset would test whether the 11-point effect survives the non-representative 203-case subset.","supporting_citations":[{"cited_title":"Agent security bench (ASB): Formalizing and benchmarking attacks and defenses in LLM-based agents,","cited_arxiv_id":null,"evidence_quote":"ASB is the first benchmark used; its attacker tool names with threat-related wording are the target of the cue-altering transformation."},{"cited_title":"Mcptox: A benchmark for tool poisoning on real-world mcp servers,","cited_arxiv_id":null,"evidence_quote":"MCPTox is the second benchmark used; its neutral tool names are replaced with threat-explicit and matched-neutral names."},{"cited_title":"AgentDojo: A dynamic environment to evaluate prompt injection attacks and defenses for LLM agents,","cited_arxiv_id":null,"evidence_quote":"AgentDojo is the third benchmark used; its benign tool names are renamed to test ASR and benign utility effects."},{"cited_title":"Tool preferences in agentic llms are unreliable,","cited_arxiv_id":null,"evidence_quote":"Shows that modifying tool descriptions without changing functionality can change tool usage, motivating the claim that metadata choices can affect agent behavior."},{"cited_title":"BiasBusters: Uncovering and mitigating tool selection bias in large language models,","cited_arxiv_id":null,"evidence_quote":"Shows that tool selection depends on how well a query matches exposed tool metadata, supporting the premise that representation affects agent actions."},{"cited_title":"Tool Specifications Matter: Uncovering and Mitigating Safety Risks in AI Agents","cited_arxiv_id":"2607.29254","evidence_quote":"Establishes a direct link between representation and safety by showing that schema-formatted tool specifications can weaken refusals and increase unsafe tool execution."},{"cited_title":"Establishing best practices in building rigorous agentic benchmarks,","cited_arxiv_id":null,"evidence_quote":"The Agentic Benchmark Checklist provides the broader benchmark-validity framing that motivates auditing an agent-security benchmark as a measurement instrument."}],"lane_accuracy":{"status":"not yet measured","lane":{"model":"deepseek/deepseek-v4-flash","input_kind":null},"url":"https://pith.science/accuracy"},"review_version":1}