{"id":"0e2c1749-a533-4a79-9b4f-90759e50e2da","arxiv_id":"2610.27115","paper_version":1,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"PROVISO uses provenance tracking and alternative comparison to select safer and more useful actions for LLM agents, outperforming baselines on four benchmarks.","lead":"PROVISO is a controller for LLM agents that decides whether to execute a proposed action by tracking the provenance of each piece of information (trusted user vs. external tool output) and comparing the action against an alternative. It improves security and utility over existing defenses on four benchmark tasks, achieving the best trade-off on all of them.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek/deepseek-v4-flash","headline":"Central claim assumes trusted contract K is complete and unambiguous; sensitivity to incomplete K not tested, so claim may not hold under realistic specification gaps.","rationale":"The reader's verdict is ACCEPT with high confidence, and the paper's evidence is solid for the given benchmarks. However, the weakest assumption identified by the reader – completeness of K – is indeed the most load-bearing. The paper does not address it experimentally, relying on benchmark-provided K which is fully specified. In practice, users and deployments may provide incomplete or ambiguous contracts, and the controller's performance under those conditions is crucial for the claimed superiority. This concern does not invalidate the presented results, but it limits their generalizability. It does not change the verdict on the paper as presented, because the paper honestly describes its setup and limitations. Therefore, verdict_should_be = UNCHANGED. I agree with the reader's identification of the weakest assumption. The paper's self-stated limitations acknowledge the benchmark-specific nature and lack of transfer, but do not explicitly test K incompleteness. This gap is significant but does not undermine the core contribution.","tokens_in":30657,"tokens_out":5018,"duration_ms":58567,"concrete_test":"For each of the four tasks, construct a test set where the trusted observation K is systematically perturbed: (a) remove one required constraint (e.g., omit the authorized recipient from the user request on ToolSafe), (b) rephrase K with underspecified details (e.g., 'pay the bill' without amount or recipient), and (c) add an ambiguous directive that could be interpreted as authorizing multiple operations. Run the same PROVISO and the top three baselines per task. Measure the relative change in F1/utility/ASR. If PROVISO's advantage over the strongest baseline is no longer statistically significant (paired test, p>0.05) on any of the perturbed sets, then the central claim is conditional on complete K and the risk is real.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's formulation (Section 3.1) defines admissible actions relative to trusted observations K = (O, S, R, T). The controller learns to value actions from outcomes, but K is taken as given. All four benchmarks provide well-specified K; the paper does not evaluate cases where K is incomplete or ambiguous. In real deployments, user requests may omit necessary constraints (e.g., 'pay the invoice' without specifying which account), or the deployment policy may be underspecified. Under such gaps, PROVISO's learned mapping may fail: it could block benign actions that rely on implicit authorization (high false positive) or allow attacks that exploit missing coverage (high false negative). Baselines that use static rules or LLM reasoning might be more robust to missing specification because they rely on general world knowledge. Therefore, the empirical superiority of PROVISO on these four tasks may not transfer to realistic settings where K is imperfect. This is a load-bearing assumption because the entire admissibility rule (and the controller's learning target) depends on K being correct and complete. The paper does not test this sensitivity, leaving the robustness of the central claim unexamined.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek/deepseek-v4-flash","summary":"The paper presents PROVISO, a controller that decides which action an LLM agent commits by separating trusted observations (user request, policy, tool interface) from external observations (tool outputs, documents, skills) and comparing the proposed action against an alternative. The controller encodes observations into provenance-typed memory, uses role-query deliberation to compare candidates, and learns to value actions from outcomes with a calibrated decision threshold. Experiments on four agent security tasks (malicious skill detection, agent misuse, unsafe tool invocation, and indirect prompt injection) show that on every task no baseline exceeds PROVISO in both security and utility, with F1 gains of up to 13.84 points over the strongest baseline. Additional skill detection benchmarks and ablation analyses confirm the contribution of the provenance separation, evidence selection, and repeated retrieval components.","tokens_in":31012,"tokens_out":1535,"duration_ms":18887,"significance":"If the empirical results hold as reported, PROVISO offers a principled solution to a key challenge in LLM agent safety: making action decisions that depend on the provenance of content rather than judging actions in isolation. The unified formulation across admission and candidate selection, together with the learnable architecture, could provide a foundation for practical guardrails. The thorough empirical evaluation across four diverse benchmarks with multiple baselines and component analyses lends credibility to the claims. The paper also releases code, supporting reproducibility.","major_comments":[{"comment":"The entire admissibility rule and the controller's learning target depend on the trusted contract K being complete and correctly specified (Section 3.1, 'We say an operation or target is licensed by K when the task contract permits it'). The paper acknowledges that PROVISO is fitted per benchmark and that attacks adapted to it are not evaluated, but it does not discuss the sensitivity to incomplete or ambiguous K. In real deployments, user requests may omit necessary constraints or the deployment policy may be underspecified. Under such gaps, PROVISO's learned mapping may fail, while static rules or LLM reasoning might be more robust. The paper should add a discussion (or experiment) on how the decision degrades when K is missing constraints or contains ambiguities, and explicitly state the assumption of K completeness as a limitation.","section":"Section 3.1, Limitations (third paragraph of Limitations section)","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The claim that PROVISO 'reduces H&R to zero while completing all benign tasks' on ANCHOR is presented as a headline result, but the benign completion is measured on a fixed panel of 80 tasks. The paper states that AGrail completes 35/80 (43.75%) on the same panel, but it does not report whether the benign tasks are representative or if the 80 tasks are a complete set from the benchmark. The authors should clarify whether the benign completion metric is based on the full ANCHOR benign test set or a subset, and whether the comparison with AGrail is statistically significant (e.g., via a paired test).","section":"Section 5.1, Table 2 (ANCHOR agent misuse)","authors_acknowledge":null,"authors_acknowledge_p":null}],"minor_comments":[{"comment":"The notation for role embeddings e_r in Eq. (3) is introduced but the dimension of e_r is not stated. Please specify the dimension (presumably 384, matching the projection width).","section":"Section 3.2, Eq. (3) and (4)","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The list of baselines for each task is thorough, but it would be helpful to state explicitly which baselines use training data (e.g., fine-tuned models) and which use frozen LLMs or static rules. Table 8 partially addresses this, but the text could cross-reference it more clearly.","section":"Section 4, 'Baselines' paragraph","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The caption says 'Paired 95% confidence intervals are in Table 15', but Table 15 is in Appendix F. Please reference the appendix table explicitly (e.g., 'Appendix F, Table 15').","section":"Section 5.3, Figure 4 caption","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The case study would be more informative if it showed the input to PROVISO (K, G, P0, P1) explicitly, not just the outcome. The provided text describes the context but the reader must infer the trust/external split.","section":"Appendix I, Case I.1","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The paper uses many abbreviations (H&R, ASR, BCE, etc.) that are defined once and then used extensively. A glossary or an abbreviation table in an appendix would help readers.","section":"Throughout","authors_acknowledge":null,"authors_acknowledge_p":null}],"recommendation":"minor_revision"},"author_rebuttal":null,"desk_editor":{"model":"deepseek/deepseek-v4-flash","letter":"The paper’s core idea—provenance-informed action selection at the commitment point—is a genuinely new framing for LLM agent security. It moves beyond judging actions in isolation by separating trusted user input from external content and learning to compare a proposed action against an alternative. The architecture (typed memory, role-query deliberation, calibrated commitment) is sensible and the ablation study convincingly shows that the provenance tracking and the retrieval structure both matter.\n\nThe evaluation is the paper's strongest asset. Four benchmarks covering the agent lifecycle, three additional skill detection benchmarks, statistical tests, and controlled component analysis. The results are clear: on every task, no existing defense beats PROVISO on both security and utility. The margins on SkillsBench and ToolSafe are substantial.\n\nThe stress-test concern about the trusted contract K being complete is real but not a hidden flaw. The paper states in Section 3.1 that K defines the contract and that it is taken as given. The limitations section explicitly says the method requires labeled outcomes from the target setting and does not study transfer or imperfect specifications. So the authors are upfront about the scope. What is missing is a sensitivity analysis—how performance degrades when K is incomplete or ambiguous—but that is a reasonable extension, not a fatal omission. The paper does not claim robustness to imperfect K.\n\nThe other soft spots are minor: candidate selection requires restorable environments and the cost of two rollouts; adaptive attacks are not evaluated. Again, these are acknowledged.\n\nThis is a well-engineered paper for a practical security problem. The results are reproducible (code released) and the evidence supports the claims within the stated scope. It does not reshape a field, but it provides a solid, new approach that practitioners could adopt.\n\nRecommendation: Send to peer review. A reviewer should push for a discussion of K completeness and perhaps some synthetic evaluation, but the paper as it stands is a legitimate contribution.","headline":"Solid engineering contribution with thorough evaluation; the central claim holds on tested benchmarks but assumes a complete trusted contract K, which is a known limitation the authors acknowledge.","tokens_in":31353,"tokens_out":950,"duration_ms":12842},"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek/deepseek-v4-flash","headline":"PROVISO introduces provenance-informed action selection for LLM agents, showing that no existing defense matches both its security and utility across four agent lifecycle tasks.","keywords":["LLM agent security","provenance-aware defense","action selection","indirect prompt injection","malicious skill detection","agent misuse","unsafe tool invocation","calibrated commitment"],"falsifier":"If any published baseline, on the same benchmarks and under identical conditions, simultaneously achieves higher security (e.g., lower attack-success rate, higher recall on unsafe actions) and higher utility (e.g., higher benign-task completion, lower false-positive rate) than PROVISO reported in this paper, the central claim would be falsified. Concretely, if TS-Guard on ToolSafe achieved F1 above 99.12% with equal or better benign retention, or if SkillGate on SkillsBench exceeded 85.81 F1 with FPR below 1.27%, the paper's strongest claim would not hold.","tokens_in":32220,"feed_emoji":"🛡️","tokens_out":3753,"duration_ms":44017,"temperature":0.7,"pith_summary":"PROVISO is a controller that decides which action an LLM agent should commit, placed between the agent's policy and the executor. It addresses two deficiencies of existing defenses: they ignore the provenance of action content (whether it comes from the trusted user or from external tool outputs) and they judge actions in isolation without comparing them to a feasible alternative. PROVISO encodes observations by provenance type, uses role-specific queries to deliberate over them, and learns the value of each candidate action from outcomes, then calibrates an operating point. Across four agent security tasks—malicious skill detection, agent misuse, unsafe tool invocation, and indirect prompt injection—no published baseline surpasses PROVISO in both security and utility. For example, PROVISO improves F1 by 13.84 points on SkillsBench and by 7.07 points on ToolSafe over the strongest baseline.","feed_headline":"Provenance-informed action selection beats all agent defenses","feed_subtitle":"PROVISO tracks where action content comes from, compares alternatives, and learns the best trade-off between security and utility.","key_machinery":"Three components: (i) provenance-typed memory, which projects each observation segment into a learned role embedding (K, G, P0, or P1) so that identical text from different sources produces distinct representations; (ii) role-query deliberation, where five learned query states (four role queries plus a commitment state) pass through L blocks of cross-attention, self-attention, and feed-forward updates to retrieve, compare, and integrate evidence; (iii) calibrated commitment, which estimates action value from the deliberation states and applies per-action offsets selected on development data to choose the best candidate.","core_discovery":"PROVISO introduces a unified formulation for agent security decisions at the commitment point: separate trusted observations (K, from user and deployment) from external observations (G, from tool outputs and skills), and choose between the proposed action P0 and an alternative P1 by comparing learned calibrated values. Its architecture—provenance-typed memory, role-query deliberation with five queries that retrieve and compare across memory, and calibrated commitment trained on native outcomes—achieves both high security and high utility across benchmarks spanning skill installation to indirect prompt injection. The paper establishes that provenance-agnostic judgment and isolation without alternatives are the root causes of the security-utility trade-off in existing defenses.","pith_inferences":["PROVISO requires labeled outcome data for each deployment and, in candidate-selection mode, sandboxed rollouts of both candidates before commitment. This limits its applicability in environments where rollouts are expensive or unavailable, a limitation the authors acknowledge.","The paper evaluates only attacks present in released benchmarks; adaptive attacks designed specifically to bypass PROVISO's provenance-based value estimation are not studied, leaving open how robust the method would be against adversaries who craft content to manipulate the learned queries.","If PROVISO's provenance-typed memory and calibrated commitment prove effective, the same pattern—separating trusted from external sources, reasoning over alternatives, and tuning an operating point from data—could generalize to other domains where an agent must choose among actions with different authority and risk profiles."],"forward_implications":["PROVISO can be deployed in any LLM agent pipeline that has a commitment point, from skill installation to final action execution, replacing handcrafted rules with learned value comparison.","The unified formulation reduces the need for separate defenses for different attack stages; the same controller architecture handles skill detection, misuse, unsafe calls, and indirect injection with only per-deployment weight fitting.","Because PROVISO compares the proposed action with an alternative (refusal or recovery), it avoids the forced trade-off between blocking attacks and completing benign tasks that plagues threshold-based guards.","On the four agent-security benchmarks, PROVISO sets new state-of-the-art results: zero Harm & Risk score on ANCHOR, 99.12 F1 on ToolSafe, 85.81 F1 on SkillsBench, and attack success rate of only 0.28% on AgentDojo while maintaining high utility."],"supporting_citations":[{"why":"Provides the ReAct framework for LLM agents as observation-action loops, which PROVISO builds upon as the agent workflow.","marker":"Yao et al. 2023"},{"why":"Supplies the SkillsBench benchmark and the SkillGate baseline for malicious skill detection, used as the primary task and a strong comparator.","marker":"Yang et al. 2026"},{"why":"Supplies the ToolSafe benchmark and the TS-Guard baseline for unsafe tool invocation, a key evaluation task.","marker":"Mou et al. 2026"},{"why":"Supplies the AgentDojo benchmark for indirect prompt injection, used in the candidate-selection evaluation.","marker":"Debenedetti et al. 2024"},{"why":"Defines the problem of indirect prompt injection, the threat model PROVISO addresses in the AgentDojo task.","marker":"Greshake et al. 2023"},{"why":"Supplies the DRIFT defense, which PROVISO uses as the alternative candidate (P1) in the AgentDojo evaluation.","marker":"Li et al. 2025a"},{"why":"Supplies the BIPIA defense, whose boundary-aware prompting defines the default candidate (P0) in AgentDojo.","marker":"Yi et al. 2025"}],"fun_headline_variants":["Provenance-based action selection tops all agent defenses","PROVISO beats every baseline on security and utility","New controller tracks action origins, picks best action","Agent security: comparing alternatives beats isolated judgment","Provenance-informed commitment ends security-utility trade-off"],"cache_read_input_tokens":0,"weakest_assumption_plain":"The trusted contract K (the user request and deployment scope) is assumed to be complete and correctly specified; if K is incomplete or ambiguous, PROVISO may block benign actions that rely on implicit authorization or fail to detect attacks that exploit missing specifications.","fun_headline_variants_meta":{"raw":{"variants":["Provenance-based action selection tops all agent defenses","PROVISO beats every baseline on security and utility","New controller tracks action origins, picks best action","Agent security: comparing alternatives beats isolated judgment","Provenance-informed commitment ends security-utility trade-off"]},"model":"deepseek/deepseek-v4-flash","effort":"low","cost_usd":5.1e-05,"raw_usage":{"cost":5.0995392e-05,"is_byok":false,"cost_details":{"upstream_inference_cost":5.0995392e-05,"upstream_inference_prompt_cost":2.7927872e-05,"upstream_inference_completions_cost":2.306752e-05},"total_tokens":1132,"prompt_tokens":861,"completion_tokens":271,"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":196}},"tokens_in":861,"tokens_out":271,"cost_source":"openrouter_usage","duration_ms":3623,"image_pages":[],"temperature":1.0,"billed_attempts":1,"reasoning_tokens":196,"transport_provider":"openrouter","cache_read_input_tokens":0,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-10-21T15:21:32.722965+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":"If any published baseline, on the same benchmarks and under identical conditions, simultaneously achieves higher security (e.g., lower attack-success rate, higher recall on unsafe actions) and higher utility (e.g., higher benign-task completion, lower false-positive rate) than PROVISO reported in this paper, the central claim would be falsified. Concretely, if TS-Guard on ToolSafe achieved F1 above 99.12% with equal or better benign retention, or if SkillGate on SkillsBench exceeded 85.81 F1 with FPR below 1.27%, the paper's strongest claim would not hold.","supporting_citations":[{"cited_title":"2023 , url =","cited_arxiv_id":null,"evidence_quote":"Provides the ReAct framework for LLM agents as observation-action loops, which PROVISO builds upon as the agent workflow."},{"cited_title":"ToolSafe: Enhancing Tool Invocation Safety of LLM-based agents via Proactive Step-level Guardrail and Feedback","cited_arxiv_id":null,"evidence_quote":"Supplies the ToolSafe benchmark and the TS-Guard baseline for unsafe tool invocation, a key evaluation task."},{"cited_title":"Advances in Neural Information Processing Systems 37 , pages =","cited_arxiv_id":null,"evidence_quote":"Supplies the AgentDojo benchmark for indirect prompt injection, used in the candidate-selection evaluation."}],"lane_accuracy":{"status":"not yet measured","lane":{"model":"deepseek/deepseek-v4-flash","input_kind":null},"url":"https://pith.science/accuracy"},"review_version":1}