Pith. sign in

REVIEW 3 major objections 4 minor 75 references

Two small open-weight models in evidence-grounded debate reach 35.30% on malware report QA, beating the best single frontier LLM (34.77%) and all cyber-specialised baselines.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-01 10:25 UTC pith:37IELZL5

load-bearing objection Useful orchestration architecture paper whose real finding is the grounding-vs-debate interaction; the 'outperform Gemini' headline rests on a three-question margin with no error bars and test-set-tuned hyperparameters. the 3 major comments →

arxiv 2607.20216 v1 pith:37IELZL5 submitted 2026-07-22 cs.CR cs.AI

Small, Free, and Effective: Orchestrating Open-Weight Small Language Models to Outperform Single LLM for Malware Analysis

Adel ElZemity , Shujun Li , Budi Arief This is my paper
classification cs.CR cs.AI
keywords small language modelsmulti-agent orchestrationadversarial debateevidence groundingmalware analysisdetonation reportsopen-weight modelsCyberSecEval benchmark
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper asks whether a small, locally deployable, open-weight system can match or beat a single large closed-weight model at reading malware detonation reports. It answers yes: a hybrid architecture that first extracts evidence from a report with retrieval tools and then has a general-purpose 4B model debate a cyber-specialised 8B model under a grounding constraint reaches 35.30% exact-match accuracy on the CyberSecEval Malware Analysis benchmark, above the 34.77% of the best ungrounded frontier model. The paper further argues the gain is architectural, not model-specific: the same evidence pipeline lifts frontier models too, and ablations trace the benefit to grounding preventing debate drift on easy retrieval questions while preserving critique gains on hard reasoning ones. The practical significance is that organisations with one consumer GPU and no API budget can get near-frontier accuracy on a security task with transparent, auditable open weights.

Core claim

On the paper's own terms, the central discovery is that orchestration can turn small open-weight models into competitive analysts: an evidence-grounded adversarial debate between Qwen3-4B and Foundation-Sec-8B, running entirely on one RTX 4090 with no API charges, answered 35.30% of 609 multiple-choice questions about Hybrid Analysis detonation reports correctly—more than the best cyber-specialised model (22.54%) and the best ungrounded frontier model, Gemini 3 Pro Preview (34.77%). When the same evidence collection was given to frontier models, Gemini reached 38.22%, showing the architecture itself, not just the small models, drives the gain. The paper attributes the improvement to two comp

What carries the argument

The load-bearing mechanism is the hybrid's two-phase 'evidence-grounded debate.' Phase 1 builds an evidence bundle E_final from the raw report by chunking, MITRE ATT&CK enrichment, sandboxed grep/jq retrieval, and embedding-similarity filtering (threshold τ=0.65). Phase 2 runs seven rounds of structured adversarial debate between a general-purpose SLM (Qwen3-4B) and a cyber-specialised model (Foundation-Sec-8B), where a validity constraint requires every claim in a response to cite a supporting snippet at similarity ≥λ=0.55—this is what blocks the drift that pure debate shows on easy questions. A verifier then checks the conclusion against the evidence bundle and falls back to the best-suppo

Load-bearing premise

The central claim hinges on the 0.53-point margin between the hybrid (35.30%) and the best ungrounded frontier model (34.77%) being a real effect rather than sampling noise, since the paper reports single runs on one 609-question test set with no error bars, no repeated runs, and no hold-out validation of the configuration choices.

What would settle it

Run the Qwen3-4B + Foundation-Sec-8B hybrid and Gemini 3 Pro (ungrounded) on the same 609 questions multiple times with different seeds and report the distribution of overall accuracy; if the 0.53-point margin disappears or flips sign under bootstrap or repeated runs, the headline superiority claim fails. Alternatively, freeze the configuration after tuning on a development split and measure accuracy on a held-out split of new detonation reports; if the hybrid no longer exceeds the frontier baseline, the margin was selection noise.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • At $0.00 API cost and roughly 6GB VRAM on a single RTX 4090, the hybrid reaches accuracy within about 3 points of grounded frontier models, making near-frontier malware-report QA locally deployable for resource-constrained teams.
  • Grounding helps frontier models too: Gemini 3 Pro climbs from 34.77% to 38.22% and Claude Opus 4.5 from 33.15% to 36.85% when run through the same evidence pipeline, so the architecture is a general reasoning benefit rather than a workaround for small models.
  • Ablations imply a design rule for multi-agent systems: pure debate drifts on easy retrieval questions, so evidence grounding is necessary to combine critique with retrieval stability; removing either component costs roughly 7–12 accuracy points on the affected difficulty tier.
  • The paper positions the result as analyst assistance, not automation: 35.30% exact-match accuracy is too low for autonomous triage, and the two-phase design gives human reviewers natural checkpoints to intervene.
  • The hybrid exceeded the best ungrounded frontier baseline on all difficulty tiers simultaneously (Easy 38.14% vs 38.00%, Medium 27.21% vs 26.00%, Hard 27.27% vs 22.73%), suggesting the benefit is not concentrated on one question type.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Because the paper reports one pass over a single 609-question set with no error bars or repeated runs, the 0.53-point margin over the best ungrounded frontier model should be read as 'comparable to' rather than 'proven better' until variance is measured.
  • The configuration (thresholds τ=0.65, λ=0.55, N=7 rounds, model pairing) was selected after seeing results on the same test set, so the reported accuracy likely overstates what a fresh deployment would achieve on new reports; a hold-out evaluation would quantify that optimism.
  • The wild-sample case studies (9/12 vs 5/12) hint that the hybrid's advantage may grow on novel evasion techniques absent from the benchmark; a larger blinded multi-annotator study on recent campaigns would test whether that advantage is real.
  • If grounding is the operative mechanism, the same architecture should transfer to other technical-artifact QA tasks—network packet captures, CVE descriptions, cloud audit logs—where evidence must be retrieved from long messy documents; that is a natural and cheap test of the paper's account.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. The paper asks whether orchestrated ensembles of open-weight small language models (SLMs), running on a single GPU, can match or exceed single large language models (LLMs) on structured question answering over malware detonation reports. The authors establish solo baselines for 11 general-purpose SLMs, 3 cyber-specialised open-weight models, and 6 frontier LLMs on the CyberSecEval / CyberSOCEval Malware Analysis benchmark (609 questions). They then design four orchestration architectures: an agentic tool-augmented pipeline, an adversarial debate framework, a hierarchical consultation system, and a hybrid that combines evidence collection with debate. The hybrid pairing Qwen3-4B with Foundation-Sec-8B reaches 35.30% overall accuracy, exceeding the best ungrounded frontier baseline (Gemini 3 Pro Preview, 34.77%) by 0.53 percentage points, while the grounded version of Gemini reaches 38.22%. Component ablations and two qualitative case studies are also reported. The paper claims that evidence-grounded orchestration substantially improves collaborative SLM performance for analyst-assistance in malware report interpretation.

Significance. If the headline comparison were statistically robust, the paper would make a useful contribution: it provides a systematic comparison of four orchestration patterns on a security benchmark, includes several thoughtful ablations, honestly reports that grounded Gemini remains superior, and releases code and traces for reproducibility. The contamination audit in Appendix C is a welcome addition. The architecture taxonomy (agentic, debate, consult, hybrid) and the identification of debate drift on easy questions are valuable and likely to inform future work. However, the central 'outperform single LLM' claim rests on a very small accuracy margin that is not supported by error bars, repeated runs, significance tests, or a hold-out tuning protocol. The paper's own ablations show nearby configurations straddling the Gemini baseline, so the headline effect should be treated as provisional until the evaluation protocol is tightened.

major comments (3)
  1. [§4.3, Table 2 and Table 3, Eq. (6)] The claim that the hybrid 'exceeds the strongest ungrounded frontier baseline' is not statistically established. The margin is 35.30% vs 34.77% on 609 questions, i.e., about 3.3 exact-match questions; the per-tier differences are 38.14% vs 38.00% (Easy, n=451, ~0.6 questions), 27.21% vs 26.00% (Medium, n=136, ~1.6 questions), and 27.27% vs 22.73% (Hard, n=22, ~1 question). No confidence intervals, repeated runs, or paired significance tests are provided, and the frontier API runs use default (nonzero) sampling. Moreover, Table 3 shows that removing the verifier alone gives 35.04%, and removing the grounding constraint gives 33.79% — values that straddle the Gemini baseline. This underscores that the reported 0.53pp advantage is within the noise of a single evaluation. The authors should add a paired significance test (e.g., McNemar or bootstrap) or repeated runs, and temper the title/cla
  2. [§3.1, §3.4, §4.1, §4.3] The headline configuration is the result of selection on the same 609-question test set, with no held-out validation. The evidence threshold τ=0.65, grounding threshold λ=0.55, the 150-keyword enrichment vocabulary, the number of debate rounds N=7, the model pair (Qwen3-4B + Foundation-Sec-8B), and the partner-selection choices in Appendix E are all described as determined empirically from performance on this benchmark. This makes the reported point estimates optimistically biased and the 'outperform' claim particularly fragile. The authors should either introduce a separate tuning split or report nested cross-validated selection, or at minimum clearly label the headline result as the best among configurations evaluated on the test set rather than a validated predictor of future performance.
  3. [Title, Abstract, §4.5 Table 4] The title and abstract claim that the SLM hybrid 'outperforms single LLM', but this is true only for the ungrounded single-pass baseline. When the same evidence pipeline is provided to frontier LLMs (Table 4), grounded Gemini achieves 38.22% and grounded Claude 36.85%, both above the SLM hybrid's 35.30%. The authors do acknowledge this in the body, but the unqualified title and the initial 'outperform single LLM' framing are misleading. The framing should be revised to state explicitly that the comparison is against ungrounded single LLMs, or that the SLM hybrid narrows the gap but does not exceed an evidence-matched frontier LLM.
minor comments (4)
  1. [Abstract vs. §3 and §4.1] The benchmark identity is inconsistent: the abstract and §3 refer to 'Meta's CyberSecEval Malware Analysis benchmark', while §4.1 calls it 'CyberSecEval Malware Analysis benchmark (CyberSOCEval test suite)' and cites reference [13] (CyberSOCEval), not [65] (CYBERSECEVAL 3). Please reconcile the naming and citations.
  2. [Table 4] Table 4 reports Easy, Hard, and Overall but omits Medium, whereas all other tables include Medium. Including Medium would make the stratified comparison complete.
  3. [§3.4, Eq. (4)-(5)] The debate is described as a Markov process and formalised with conditional notation, but the number of rounds N is not defined in the formalism; N appears only later in Table 2. A brief definition would improve precision.
  4. [§4.6 and Appendix D] The case-study evaluation is acknowledged as preliminary, and the authors note the single-evaluator limitation. It would be helpful to state whether the ground-truth labels and the blinded outputs are released in the repository in a machine-readable form, since the current text says they are 'provided' but does not specify the format.

Circularity Check

0 steps flagged

No circular derivation: the hybrid accuracy is an external-benchmark measurement, not a quantity defined by its fitted inputs.

full rationale

The paper's central claim—that the Qwen3-4B + Foundation-Sec-8B hybrid reaches 35.30% on CyberSecEval Malware Analysis—is an empirical measurement against an external benchmark, not a quantity derived by construction from the system's own equations or from an author-imported uniqueness theorem. Hyperparameters such as τ=0.65, λ=0.55, the 150-keyword vocabulary, the number of debate rounds, and the model pairing were indeed chosen empirically on the same 609-question test set (e.g., §3.1 'a threshold determined empirically', §4.1 'selected based on its leading performance across all solo and orchestrated configurations'), which creates a real statistical-selection concern: the reported 0.53pp margin over Gemini 3 Pro Preview has no error bars, repeated runs, or hold-out validation, so the point estimate is optimistically biased and the 'outperform' claim is not statistically established. However, this is a correctness/robustness limitation, not circularity: the final accuracy is not definitionally equal to the selection criterion, and the evaluation is anchored to an external benchmark rather than to a self-citation chain or an input-output tautology. The only author-overlapping citation is [75] (Çetin, Ekmekcioglu, Arief, Hernandez-Castro), used as non-load-bearing related work on LLMs for static code analysis; it does not support the paper's main result. Consequently, no circular step meeting the required evidentiary bar can be quoted, and the appropriate score is 2 for the presence of a minor, non-load-bearing self-citation, with the central derivation otherwise independent.

Axiom & Free-Parameter Ledger

7 free parameters · 5 axioms · 0 invented entities

The headline result depends on at least seven free parameters that were tuned on the same benchmark used for the reported accuracy, plus several domain assumptions about benchmark validity and contamination. This is the main reason the reported 35.30% should be treated as a fitted maximum, not an out-of-sample prediction.

free parameters (7)
  • evidence threshold τ = 0.65
    Cosine-similarity threshold for retaining evidence snippets; 'determined empirically to balance evidence recall against noise' (§3.1). Affects all agentic/hybrid results.
  • grounding constraint λ = 0.55
    Minimum cosine similarity between debate claims and evidence snippets; 'set lower than τ' since claims are paraphrases (§3.4). Tuned on the benchmark.
  • chunk size / overlap = 512 tokens / 64 tokens
    Ingestion chunking parameters in the agentic pipeline (§3.1); chosen by the authors.
  • enrichment keyword vocabulary size = 150
    Curated list of technique-indicative keywords; 'size determined empirically to balance enrichment coverage against false trigger rate' (§3.1).
  • number of debate rounds N = 7
    Selected as optimal trade-off between hard-question gains and easy-question degradation (Figure 4, Table 2).
  • model pair selection = Qwen3-4B + Foundation-Sec-8B
    Chosen because Qwen3-4B had 'leading performance across all solo and orchestrated configurations' and Foundation-Sec-8B provided cyber expertise (§4.1). Selected from models tested on the same benchmark.
  • ψ extraction minimum length = 10 words
    Heuristic for extracting sub-question in the consult system when no explicit marker is found (§3.3).
axioms (5)
  • domain assumption Accuracies on the 609-question CyberSOCEval/CyberSecEval multiple-choice benchmark are stable point estimates that support ranking systems differing by <1 percentage point.
    The central claim depends on comparing point estimates without error bars (§4.2–4.5).
  • domain assumption The evaluated models (including unaudited frontier models) have no meaningful test-set contamination that would favor the SLM ensembles.
    Contamination audit was performed only for Foundation-Sec-8B (Appendix C); no audit for Qwen3-4B or frontier models.
  • domain assumption Cosine similarity of all-MiniLM-L6-v2 embeddings is a valid measure of evidence relevance and claim grounding.
    Used for evidence filtering (τ) and claim validation (λ) (§3.1, §3.4).
  • standard math The difficulty-tier split (Easy 74% of questions) is accepted as given and not a source of bias in the weighted overall accuracy.
    Weighting formula (6) is standard; the dominance of Easy questions is inherent to the benchmark.
  • domain assumption Single-evaluator scoring of the 12 wild samples yields reliable ground-truth classification.
    Appendix D says a single author with 3+ years of malware analysis experience scored blinded outputs; no inter-rater reliability.

pith-pipeline@v1.3.0-alltime-deepseek · 24013 in / 13803 out tokens · 108429 ms · 2026-08-01T10:25:57.115526+00:00 · methodology

0 comments
read the original abstract

Malware analysis demands rapid interpretation of complex detonation reports spanning filesystem, network, and process behaviours. While large language models (LLMs) demonstrate impressive capabilities for technical artifact interpretation, the opacity and escalating API costs of closed-weight frontier models motivate exploration of open-weight alternatives. However, many open-weight models are large, demanding significant compute resources and incurring non-trivial hosting costs that place them beyond reach for resource-constrained deployments. This paper investigates whether orchestrated ensembles of small language models (SLMs) can match or exceed single LLM performance on structured questions about malware detonation reports. We established baselines by testing eleven open-weight SLMs, three cyber security pre-trained models, and six frontier LLMs on Meta's CyberSecEval Malware Analysis benchmark. We then designed and evaluated four orchestration architectures: (i) a multi-agent pipeline that decomposes analysis into structured evidence-collection and reasoning stages, (ii) an adversarial debate framework in which two agents iteratively critique each other's reasoning, (iii) a hierarchical consultation system that pairs a general-purpose SLM with a cyber-specialised expert model, and (iv) a hybrid architecture that combines evidence-grounded pipelines with adversarial debate reasoning. The hybrid system (Qwen3-4B with Foundation-Sec-8B) achieved 35.30% overall accuracy, exceeding the strongest cyber-specialised baseline (22.54%) and the strongest ungrounded frontier baseline (34.77%); when given the same evidence pipeline, grounded Gemini remained the strongest configuration at 38.22%. These findings show that evidence-grounded orchestration can substantially improve the performance of collaborative SLMs for supporting interpretation of malware detonation reports.

Figures

Figures reproduced from arXiv: 2607.20216 by Adel ElZemity, Budi Arief, Shujun Li.

Figure 1
Figure 1. Figure 1: Hybrid orchestration architecture. Left (evidence collection phase): The four-stage agentic pipeline shared with the standalone agentic system (Section 3.1)— ingestion, enrichment, tool-search, and evidence mining—extracts and validates sup￾porting evidence. The standalone agentic system terminates after a single reasoning and verifier step using one model. Right (debate reasoning phase): Unique to the hyb… view at source ↗
Figure 2
Figure 2. Figure 2: Consult system architecture. The tested agent (MA, general-purpose SLM) receives the full benchmark prompt qtask and iteratively consults the stateless expert agent (ME, cyber-specialised), which never sees qtask. Each round, ψ extracts a sub￾question q (t) sub from the tested agent’s rationale; the expert returns focused domain guidance et, which accumulates in HE across N rounds. The tested agent then pr… view at source ↗
Figure 3
Figure 3. Figure 3: The ablation analysis of the agentic system showing the impact of command￾line tools across four representative SLMs. Tool access provides the largest performance boost for all models. Qwen3-4B with tools achieves the best overall performance among agentic configurations (25.11%), surpassing all open-weight SLM solo baselines. 4.3 Orchestrated Systems Performance [PITH_FULL_IMAGE:figures/full_fig_p014_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: The debate system’s performance as a function of debate rounds (1–15 rounds), where each SLM debates with Foundation-Sec-8B. Results show consistent improve￾ment on hard questions (solid lines) that plateaus after 7–10 rounds, while easy ques￾tions (dashed lines) exhibit clear degradation with increased rounds, declining from peak performance at rounds 7–10 to lower accuracy by round 15. Qwen3-4B achieves … view at source ↗
Figure 5
Figure 5. Figure 5: The consult system’s performance comparison where each SLM consults Foundation-Sec-8B (cyber 8B) versus a general LLM or a larger 70B cyber expert. Cyber-specialised experts consistently outperform general LLMs, but parameter scal￾ing from 8B to 70B yields minimal additional gains. Qwen3-4B achieves the highest consult performance (22.74% overall with Foundation-Sec-8B). Hybrid System Results The hybrid sy… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

75 extracted references · 8 canonical work pages

  1. [1]

    Sensors23(2), 946:1–946:17 (2023).https://doi.org/10.3390/s23020 946

    Akhtar, M.S., Feng, T.: Evaluation of machine learning algorithms for malware detection. Sensors23(2), 946:1–946:17 (2023).https://doi.org/10.3390/s23020 946

    work page doi:10.3390/s23020 2023

  2. [2]

    Preprint arXiv:2409.07587 [cs.CR] (2024).https://doi.org/10.48550/arXiv.2409.07587

    Al-Karaki, J., Khan, M.A.Z., Omar, M.: Exploring LLMs for malware de- tection: Review, framework design, and countermeasure approaches. Preprint arXiv:2409.07587 [cs.CR] (2024).https://doi.org/10.48550/arXiv.2409.07587

    Pith/arXiv arXiv doi:10.48550/arxiv.2409.07587 2024

  3. [3]

    Online document (2025),https://ww w-cdn.anthropic.com/bf10f64990cfda0ba858290be7b8cc6317685f47.pdf

    Anthropic: System card: Claude Opus 4.5. Online document (2025),https://ww w-cdn.anthropic.com/bf10f64990cfda0ba858290be7b8cc6317685f47.pdf

    2025

  4. [4]

    IEEE Access9, 83252–83271 (2021)

    Aslan, Ö., Ozkan-okay, M., Gupta, D.: Intelligent behavior-based malware detec- tion system on cloud computing environment. IEEE Access9, 83252–83271 (2021). https://doi.org/10.1109/access.2021.3087316

    arXiv 2021

  5. [5]

    IEEE Access8, 14329– 14343 (2020).https://doi.org/10.1109/access.2020.2966321

    Belaoued, M., Derhab, A., Mazouzi, S., Khan, F.A.: MACoMal: A multi-agent based collaborative mechanism for anti-malware assistance. IEEE Access8, 14329– 14343 (2020).https://doi.org/10.1109/access.2020.2966321

    arXiv 2020

  6. [6]

    Preprint arXiv:2506.02153 (2025).https://doi.org/10.48550/arXiv.2506.02153

    Belcak, P., Heinrich, G., Diao, S., Fu, Y., Dong, X., Muralidharan, S., Lin, Y.C., Molchanov, P.: Small language models are the future of agentic AI. Preprint arXiv:2506.02153 (2025).https://doi.org/10.48550/arXiv.2506.02153

    Pith/arXiv arXiv doi:10.48550/arxiv.2506.02153 2025

  7. [7]

    Digital Camouflage

    Böke, E., Torka, S.: “Digital Camouflage”: The LLVM challenge in LLM-based malwaredetection.JournalofSystemsandSoftwarepp.112646:1–112646:11(2025)

    2025

  8. [8]

    Cai, W., Jiang, J., Wang, F., Tang, J., Kim, S., Huang, J.: A survey on mixture of expertsinlargelanguagemodels.IEEETrans.onKnowledgeandDataEngineering 37, 3896–3915 (2024).https://doi.org/10.1109/tkde.2025.3554028

    arXiv 2024

  9. [9]

    In: Proceedings of the 30th USENIX Se- curity Symposium

    Carlini, N., Tramer, F., Wallace, E., Jagielski, M., Herbert-Voss, A., Lee, K., Roberts, A., Brown, T., Song, D., Erlingsson, U., Oprea, A., Raffel, C.: Extracting training data from large language models. In: Proceedings of the 30th USENIX Se- curity Symposium. pp. 2633–2650. USENIX Association (2021),https://www.us enix.org/conference/usenixsecurity21/p...

    2021

  10. [10]

    PeerJ Computer Science 6, e285:1–e285:23 (2020).https://doi.org/10.7717/peerj-cs.285

    Catak, F.O., Yazi, A.F., Elezaj, O., Ahmed, J.: Deep learning based sequential model for malware analysis using windows exe API calls. PeerJ Computer Science 6, e285:1–e285:23 (2020).https://doi.org/10.7717/peerj-cs.285

    work page doi:10.7717/peerj-cs.285 2020

  11. [11]

    Preprint arXiv:2309.13007 [cs.CL] (2023).https://doi.org/10.48550/arXiv.2309.13007

    Chen, J.C.Y., Saha, S., Bansal, M.: ReConcile: Round-table conference improves reasoning via consensus among diverse LLMs. Preprint arXiv:2309.13007 [cs.CL] (2023).https://doi.org/10.48550/arXiv.2309.13007

    Pith/arXiv arXiv doi:10.48550/arxiv.2309.13007 2023

  12. [12]

    Website (2024),https://www.hybrid-analysis.com/, accessed 2025-12-01

    CrowdStrike: Hybrid Analysis: Free automated malware analysis service. Website (2024),https://www.hybrid-analysis.com/, accessed 2025-12-01

    2024

  13. [13]

    ElZemity et al

    Deason, L., Bali, A., Bejean, C., Bolocan, D., Crnkovich, J., Croitoru, I., Durai, K., Midler, C., Miron, C., Molnar, D., Moon, B., Ostarcevic, B., Peltea, A., Rosenberg, 28 A. ElZemity et al. M., Sandu, C., Saputkin, A., Shah, S., Stan, D., Szocs, E., Wan, S., Whitman, S., Krasser, S., Saxe, J.: CyberSOCEval: Benchmarking LLMs capabilities for mal- ware ...

    doi:10.48550/arxiv.2509.20166 2025

  14. [14]

    LLM repo (2024),https://huggingface.co/DeepHat /DeepHat-V1-7B, accessed 2025-12-01

    DeepHat: DeepHat-V1-7B. LLM repo (2024),https://huggingface.co/DeepHat /DeepHat-V1-7B, accessed 2025-12-01

    2024

  15. [15]

    LLM repo (2024),https://huggingf ace.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B, accessed 2025-12-01

    DeepSeek: DeepSeek-R1-Distill-Qwen-1.5B. LLM repo (2024),https://huggingf ace.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B, accessed 2025-12-01

    2024

  16. [16]

    Preprint arXiv:2512.02556 [cs.CL] (2025).https://doi.org/10.48550/a rXiv.2512.02556

    DeepSeek-AI Team: DeepSeek-V3.2: Pushing the frontier of open large language models. Preprint arXiv:2512.02556 [cs.CL] (2025).https://doi.org/10.48550/a rXiv.2512.02556

    Pith/arXiv arXiv doi:10.48550/arxiv.2512.02556 2025

  17. [17]

    In: Proceedings of the 41st International Conference on Machine Learning

    Du, Y., Li, S., Torralba, A., Tenenbaum, J.B., Mordatch, I.: Improving factuality and reasoning in language models through multiagent debate. In: Proceedings of the 41st International Conference on Machine Learning. pp. 8465–8479. PMLR (2025),https://raw.githubusercontent.com/mlresearch/v235/main/assets/ du24e/du24e.pdf

    2025

  18. [18]

    IEEE Transactions on Information Forensics and Security16, 838–853 (2020).https: //doi.org/10.1109/tifs.2020.3021924

    Fan, M., Wei, W., Xie, X., Liu, Y., Guan, X., Liu, T.: Can we trust your ex- planations? sanity checks for interpreters in Android malware analysis. IEEE Transactions on Information Forensics and Security16, 838–853 (2020).https: //doi.org/10.1109/tifs.2020.3021924

    arXiv 2020

  19. [19]

    LLM repo (2024),https://huggingface.co /fdtn-ai/Foundation-Sec-8B-Instruct, accessed 2025-12-01

    fdtn-ai: Foundation-Sec-8B-Instruct. LLM repo (2024),https://huggingface.co /fdtn-ai/Foundation-Sec-8B-Instruct, accessed 2025-12-01

    2024

  20. [20]

    Preprint arXiv:2312.11805v5 [cs.CL] (2025),https://arxiv.org/abs/2312.118 05v5

    Gemini Team, Google: Gemini: A family of highly capable multimodal models. Preprint arXiv:2312.11805v5 [cs.CL] (2025),https://arxiv.org/abs/2312.118 05v5

    Pith/arXiv arXiv 2025

  21. [21]

    ACM Computing Surveys58(3), 67:1–67:38 (2025).https://doi.org/10.1145/ 3747587

    Giarimpampa, D., Meier, R., Bissyande, T.F., Lenders, V., Klein, J.: Exploring the role of artificial intelligence in enhancing security operations: A systematic review. ACM Computing Surveys58(3), 67:1–67:38 (2025).https://doi.org/10.1145/ 3747587

    2025

  22. [22]

    Preprint arXiv:2308.08493v3 [cs.CL] (2024),https://arxi v.org/abs/2308.08493v3

    Golchin, S., Surdeanu, M.: Time travel in LLMs: Tracing data contamination in large language models. Preprint arXiv:2308.08493v3 [cs.CL] (2024),https://arxi v.org/abs/2308.08493v3

    Pith/arXiv arXiv 2024

  23. [23]

    LLM repo (2024),https://huggingface.co /google/gemma-3-4b-it, accessed 2025-12-01

    Google DeepMind: Gemma 3 4B IT. LLM repo (2024),https://huggingface.co /google/gemma-3-4b-it, accessed 2025-12-01

    2024

  24. [24]

    Web page (2025),https://www.group-ib.com/blog/c lickfix-the-social-engineering-technique-hackers-use-to-manipulate-v ictims/, accessed: 2026-02-04

    Group-IB Threat Intelligence: ClickFix: The social engineering technique hackers use to manipulate victims. Web page (2025),https://www.group-ib.com/blog/c lickfix-the-social-engineering-technique-hackers-use-to-manipulate-v ictims/, accessed: 2026-02-04

    2025

  25. [25]

    ACM Transactions on Software Engineering and Methodology33, 220:1–220:79 (2023).https://doi.org/10.1145/3695988

    Hou,X.,Zhao,Y.,Liu,Y.,Yang,Z.,Wang,K.,Li,L.,Luo,X.,Lo,D.,Grundy,J.C.: Large language models for software engineering: A systematic literature review. ACM Transactions on Software Engineering and Methodology33, 220:1–220:79 (2023).https://doi.org/10.1145/3695988

    doi:10.1145/3695988 2023

  26. [26]

    Model repository (2021),https://huggingfac e.co/sentence-transformers/all-MiniLM-L6-v2, accessed 2025-12-01

    Hugging Face: all-MiniLM-L6-v2. Model repository (2021),https://huggingfac e.co/sentence-transformers/all-MiniLM-L6-v2, accessed 2025-12-01

    2021

  27. [27]

    LLM repo (2024),https://huggingface.co/Hug gingFaceTB/SmolLM2-1.7B, accessed 2025-12-01

    Hugging Face: SmolLM2-1.7B. LLM repo (2024),https://huggingface.co/Hug gingFaceTB/SmolLM2-1.7B, accessed 2025-12-01

    2024

  28. [28]

    Web page (2025),https://www.huntress.com/blog/clickfix-malware-buried-in-image s, accessed: 2026-02-04 Orchestrating SLMs to Outperform Single LLM for Malware Analysis 29

    Huntress Threat Ops: ClickFix gets creative: Malware buried in images. Web page (2025),https://www.huntress.com/blog/clickfix-malware-buried-in-image s, accessed: 2026-02-04 Orchestrating SLMs to Outperform Single LLM for Malware Analysis 29

    2025

  29. [29]

    In: Proceedings of the 16th International Conference on Availability, Reli- ability and Security

    Kuppa, A., Aouad, L., Le-Khac, N.A.: Linking CVE’s to MITRE ATT&CK tech- niques. In: Proceedings of the 16th International Conference on Availability, Reli- ability and Security. pp. 21:1–21:12. ACM (2021).https://doi.org/10.1145/34 65481.3465758

    arXiv 2021

  30. [30]

    Information Systems pp

    Leon, M.: GPT-5 and open-weight large language models: Advances in reasoning, transparency, and control. Information Systems pp. 102620:1–102620:9 (2025).ht tps://doi.org/10.1016/j.is.2025.102620

    arXiv 2025

  31. [31]

    Preprint arXiv:2510.14113 [cs.CL] (2025).https://doi.org/10.48550/arXiv.2510.14113

    Levi, M., Ohayon, D., Blobstein, A., Sagi, R., Molloy, I., Allouche, Y.: Toward cybersecurity-expert small language models. Preprint arXiv:2510.14113 [cs.CL] (2025).https://doi.org/10.48550/arXiv.2510.14113

    Pith/arXiv arXiv doi:10.48550/arxiv.2510.14113 2025

  32. [32]

    Journal of Information Security and Applications95, 104284:1–104284:18 (2025)

    Li, M.Q., Fung, B.C.M.: Security concerns for large language models: A survey. Journal of Information Security and Applications95, 104284:1–104284:18 (2025). https://doi.org/10.1016/j.jisa.2025.104284

    arXiv 2025

  33. [33]

    1527–1539

    Lin, F., Kim, D.J., Chen, T.H.P.: SOEN-101: Code Generation by Emulating Soft- ware Process Models Using Large Language Model Agents, pp. 1527–1539. IEEE (2025).https://doi.org/10.1109/ICSE55347.2025.00140

    arXiv 2025

  34. [34]

    Preprint arXiv:2310.02170 [cs.CL] (2023).https://doi.org/10.48550/arXiv.2310.02170

    Liu, Z., Zhang, Y., Li, P., Liu, Y., Yang, D.: A dynamic LLM-powered agent network for task-oriented agent collaboration. Preprint arXiv:2310.02170 [cs.CL] (2023).https://doi.org/10.48550/arXiv.2310.02170

    Pith/arXiv arXiv doi:10.48550/arxiv.2310.02170 2023

  35. [35]

    Preprint arXiv:2409.15790 [cs.CL] (2025).https://doi.org/10.48550/arXiv.2409.15790

    Lu, Z., Li, X., Cai, D., Yi, R., Liu, F., Zhang, X., Lane, N.D., Xu, M.: Small language models: Survey, measurements, and insights. Preprint arXiv:2409.15790 [cs.CL] (2025).https://doi.org/10.48550/arXiv.2409.15790

    Pith/arXiv arXiv doi:10.48550/arxiv.2409.15790 2025

  36. [36]

    IEEE Access12,68624–68636(2024).https://doi.org/10.1109/access.2024.3400167

    Mahmoud, R., Anagnostopoulos, M., Pastrana, S., Pedersen, J.M.: Redefining mal- ware sandboxing: Enhancing analysis through sysmon and ELK integration. IEEE Access12,68624–68636(2024).https://doi.org/10.1109/access.2024.3400167

    arXiv 2024

  37. [37]

    Web page (2025),https://cloud.google.com/blog/topics/threat-intelligence/unc514 2-etherhiding-distribute-malware, accessed: 2026-02-04

    Mandiant Threat Intelligence: New group on the block: UNC5142 leverages Ether- Hiding to distribute malware, Google Cloud Threat Intelligence Blog. Web page (2025),https://cloud.google.com/blog/topics/threat-intelligence/unc514 2-etherhiding-distribute-malware, accessed: 2026-02-04

    2025

  38. [38]

    IEEE Access13, 61611–61640 (2024).https://doi.org/10.1109/access.2025.3555926

    Manthena, H., Shajarian, S., Kimmell, J., Abdelsalam, M., Khorsandroo, S., Gupta, M.: Explainable artificial intelligence (XAI) for malware analysis: A survey of techniques, applications, and open challenges. IEEE Access13, 61611–61640 (2024).https://doi.org/10.1109/access.2025.3555926

    arXiv 2024

  39. [39]

    Journal of Cyber- security10(1), tyad023:1–tyad023:18 (2024).https://doi.org/10.1093/cybsec /tyad023

    Mat, N., Jamil, N., Yusoff, Y., Kiah, M.L.M.: A systematic literature review on advanced persistent threat behaviors and its detection strategy. Journal of Cyber- security10(1), tyad023:1–tyad023:18 (2024).https://doi.org/10.1093/cybsec /tyad023

    work page doi:10.1093/cybsec 2024

  40. [40]

    LLM repo (2024),https://huggingface.co/m eta-llama/Llama-3.1-8B-Instruct, accessed 2025-12-01

    Meta AI: Llama 3.1 8B Instruct. LLM repo (2024),https://huggingface.co/m eta-llama/Llama-3.1-8B-Instruct, accessed 2025-12-01

    2024

  41. [41]

    LLM repo (2024),https://huggingface.co/meta-llama /Llama-3.2-1B, accessed 2025-12-01

    Meta AI: Llama 3.2 1B. LLM repo (2024),https://huggingface.co/meta-llama /Llama-3.2-1B, accessed 2025-12-01

    2024

  42. [42]

    Web page (2025),https://ai.meta.com/blog/llama-4-multimo dal-intelligence

    Meta Team: The Llama 4 herd: The beginning of a new era of natively multimodal AI innovation. Web page (2025),https://ai.meta.com/blog/llama-4-multimo dal-intelligence

    2025

  43. [43]

    LLM repo (2024),https://huggingface.co/m icrosoft/Phi-3.5-mini-instruct, accessed 2025-12-01

    Microsoft: Phi-3.5 Mini Instruct. LLM repo (2024),https://huggingface.co/m icrosoft/Phi-3.5-mini-instruct, accessed 2025-12-01

    2024

  44. [44]

    Microsoft Threat Intelligence: Lumma Stealer: Breaking down the delivery tech- niques and capabilities of a prolific infostealer. Web page (2025),https://www.mi crosoft.com/en-us/security/blog/2025/05/21/lumma-stealer-breaking-dow n-the-delivery-techniques-and-capabilities-of-a-prolific-infostealer, accessed: 2026-02-04 30 A. ElZemity et al

    2025

  45. [45]

    IEEE Access10, 99129–99149 (2022).https://doi.org/ 10.1109/access.2022.3207287

    Mienye, I.D., Sun, Y.: A survey of ensemble learning: Concepts, algorithms, appli- cations, and prospects. IEEE Access10, 99129–99149 (2022).https://doi.org/ 10.1109/access.2022.3207287

    arXiv 2022

  46. [46]

    LLM repo (2024),https://huggingface

    Mistral AI: Ministral-8B-Instruct-2410. LLM repo (2024),https://huggingface. co/mistralai/Ministral-8B-Instruct-2410, accessed 2025-12-01

    2024

  47. [47]

    Preprint arXiv:2303.08774v6 [cs.CL] (2024),https://arxiv.org/abs/2303.08774v6

    OpenAI Team: GPT-4 technical report. Preprint arXiv:2303.08774v6 [cs.CL] (2024),https://arxiv.org/abs/2303.08774v6

    Pith/arXiv arXiv 2024

  48. [48]

    ACM Computing Surveys52, 88:1–88:48 (2019).https://doi.org/10.1145/3329786

    Or-Meir, O., Nissim, N., Elovici, Y., Rokach, L.: Dynamic malware analysis in the modern era—a state of the art survey. ACM Computing Surveys52, 88:1–88:48 (2019).https://doi.org/10.1145/3329786

    doi:10.1145/3329786 2019

  49. [49]

    Expert Systems with Applications256, 124912:1–124912:13 (2024).https://doi.org/10.1016/j.eswa.2024.124912

    Patsakis, C., Casino, F., Lykousas, N.: Assessing LLMs in malicious code deobfus- cation of real-world malware campaigns. Expert Systems with Applications256, 124912:1–124912:13 (2024).https://doi.org/10.1016/j.eswa.2024.124912

    arXiv 2024

  50. [50]

    Web page (2025),https://www.picussecurity.com/resource/blo g/etherhiding-how-web3-infrastructure-enables-stealthy-malware-distr ibution, accessed: 2026-02-04

    Picus Security: EtherHiding: How Web3 infrastructure enables stealthy malware distribution. Web page (2025),https://www.picussecurity.com/resource/blo g/etherhiding-how-web3-infrastructure-enables-stealthy-malware-distr ibution, accessed: 2026-02-04

    2025

  51. [51]

    IEEE Access9, 76248–76259 (2021).https://doi.org/10.1109/ac cess.2021.3067284

    Qaisar, Z.H., Almotiri, S.H., Al Ghamdi, M.A., Nagra, A.A., Ali, G.: A scalable and efficient multi-agent architecture for malware protection in data sharing over mobile cloud. IEEE Access9, 76248–76259 (2021).https://doi.org/10.1109/ac cess.2021.3067284

    arXiv 2021

  52. [52]

    LLM repo (2024),https://huggingface.co /Qwen/Qwen2.5-1.5B-Instruct, accessed 2025-12-01

    Qwen Team: Qwen2.5 1.5B Instruct. LLM repo (2024),https://huggingface.co /Qwen/Qwen2.5-1.5B-Instruct, accessed 2025-12-01

    2024

  53. [53]

    LLM repo (2024),https://huggingfac e.co/Qwen/Qwen2.5-Coder-7B-Instruct, accessed 2025-12-01

    Qwen Team: Qwen2.5-Coder 7B Instruct. LLM repo (2024),https://huggingfac e.co/Qwen/Qwen2.5-Coder-7B-Instruct, accessed 2025-12-01

    2024

  54. [54]

    LLM repo (2024),https://huggingface.co/Qwen/Qw en3-0.6B, accessed 2025-12-01

    Qwen Team: Qwen3-0.6B. LLM repo (2024),https://huggingface.co/Qwen/Qw en3-0.6B, accessed 2025-12-01

    2024

  55. [55]

    LLM repo (2025),https://huggingface.co/Qwen/Qwen 3-4B, accessed 2025-12-01

    Qwen Team: Qwen3-4B. LLM repo (2025),https://huggingface.co/Qwen/Qwen 3-4B, accessed 2025-12-01

    2025

  56. [56]

    Scientific Reports15, 13755:1–13755:23 (2025)

    Raza, M., Jahangir, Z., Riaz, M.B., Saeed, M.J., Sattar, M.A.: Industrial appli- cations of large language models. Scientific Reports15, 13755:1–13755:23 (2025). https://doi.org/10.1038/s41598-025-98483-1

    doi:10.1038/s41598-025-98483-1 2025

  57. [57]

    Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery8(4), e1249:1–e1249:18 (2018).https: //doi.org/10.1002/widm.1249

    Sagi, O., Rokach, L.: Ensemble learning: A survey. Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery8(4), e1249:1–e1249:18 (2018).https: //doi.org/10.1002/widm.1249

    doi:10.1002/widm.1249 2018

  58. [58]

    ACM Computing Surveys56, 314:1–314:40 (2024).https://doi.org/10.1145/3677374

    Saqib, M., Mahdavifar, S., Fung, B.C.M., Charland, P.: A comprehensive analysis of explainable AI for malware hunting. ACM Computing Surveys56, 314:1–314:40 (2024).https://doi.org/10.1145/3677374

    doi:10.1145/3677374 2024

  59. [59]

    Web page (2026),https://blog.s ekoia.io/meet-iclickfix-a-widespread-wordpress-targeting-framework-u sing-the-clickfix-tactic/, accessed: 2026-02-04

    Sekoia.io Threat & Detection Research: Meet IClickFix: a widespread WordPress- targeting framework using the ClickFix tactic. Web page (2026),https://blog.s ekoia.io/meet-iclickfix-a-widespread-wordpress-targeting-framework-u sing-the-clickfix-tactic/, accessed: 2026-02-04

    2026

  60. [60]

    SentinelOne Labs web page (2026),https://www.sentin elone.com/labs/building-an-adversarial-consensus-engine-multi-agent-l lms-for-automated-malware-analysis/, accessed: 2026-07-14

    Stokes, P.: Building an adversarial consensus engine: Multi-agent LLMs for auto- mated malware analysis. SentinelOne Labs web page (2026),https://www.sentin elone.com/labs/building-an-adversarial-consensus-engine-multi-agent-l lms-for-automated-malware-analysis/, accessed: 2026-07-14

    2026

  61. [61]

    Strom, B.E., Applebaum, A., Miller, D.P., Nickels, K.C., Pennington, A.G., Thomas, C.B.: MITRE ATT&CK: Design and philosophy. Tech. rep., The MITRE Corporation (2018),https://www.mitre.org/sites/default/files/2021-11/pr s-19-01075-28-mitre-attack-design-and-philosophy.pdf Orchestrating SLMs to Outperform Single LLM for Malware Analysis 31

    2018

  62. [62]

    Preprint arXiv:2501.06322 [cs.AI] (2025).https://doi.org/10.48550/arXiv.2501.06322

    Tran, K.T., Dao, D., Nguyen, M.D., Pham, Q.V., O’Sullivan, B., Nguyen, H.D.: Multi-agent collaboration mechanisms: A survey of LLMs. Preprint arXiv:2501.06322 [cs.AI] (2025).https://doi.org/10.48550/arXiv.2501.06322

    Pith/arXiv arXiv doi:10.48550/arxiv.2501.06322 2025

  63. [63]

    Preprint arXiv:2509.01835 [cs.CR] (2025).https://doi.org/10.48550/arXiv.2509.01835

    Ullah, S., Balasubramanian, P., Guo, W., Burnett, A., Pearce, H., Kruegel, C., Vigna, G., Stringhini, G.: From CVE entries to verifiable exploits: An automated multi-agent framework for reproducing CVEs. Preprint arXiv:2509.01835 [cs.CR] (2025).https://doi.org/10.48550/arXiv.2509.01835

    doi:10.48550/arxiv.2509.01835 2025

  64. [64]

    Scientific Reports15, 37979:1–37979:21 (2025).https://doi.org/10.1038/s41598-025-21848-z

    Uysal,D.T.,Yoo,P.D.,Taha,K.,Yeun,C.,Damiani,E.:Amulti-labelvisualisation approach for malware behaviour analysis. Scientific Reports15, 37979:1–37979:21 (2025).https://doi.org/10.1038/s41598-025-21848-z

    work page doi:10.1038/s41598-025-21848-z 2025

  65. [65]

    Preprint arXiv:2408.01605 [cs.CR] (2024)

    Wan, S., Nikolaidis, C., Song, D., Molnar, D., Crnkovich, J., Grace, J., Bhatt, M., Chennabasappa, S., Whitman, S., Ding, S., Ionescu, V., Li, Y., Saxe, J.: CYBERSECEVAL 3: Advancing the evaluation of cybersecurity risks and ca- pabilities in large language models. Preprint arXiv:2408.01605 [cs.CR] (2024). https://doi.org/10.48550/arXiv.2408.01605

    Pith/arXiv arXiv doi:10.48550/arxiv.2408.01605 2024

  66. [66]

    https://doi.org/10.48550/arXiv.2406.04692

    Wang, J., Wang, J., Athiwaratkun, B., Zhang, C., Zou, J.: Mixture-of-agents en- hanceslargelanguagemodelcapabilities.PreprintarXiv:2406.04692[cs.CL](2024). https://doi.org/10.48550/arXiv.2406.04692

    Pith/arXiv arXiv doi:10.48550/arxiv.2406.04692 2024

  67. [67]

    Preprint arXiv:2410.01257 [cs.LG] (2024).https://doi.org/10.48550/arXiv .2410.01257

    Wang, Z., Bukharin, A., Delalleau, O., Egert, D., Shen, G., Zeng, J., Kuchaiev, O., Dong, Y.: HelpSteer2-preference: Complementing ratings with preferences. Preprint arXiv:2410.01257 [cs.LG] (2024).https://doi.org/10.48550/arXiv .2410.01257

    Pith/arXiv arXiv doi:10.48550/arxiv.2410.01257 2024

  68. [68]

    Preprint arXiv:2308.08155 [cs.AI].https://doi.org/10.48550/arXiv.2308.08155

    Wu, Q., Bansal, G., Zhang, J., Wu, Y., Li, B., Zhu, E., Jiang, L., Zhang, X., Zhang, S., Liu, J., Awadallah, A.H., White, R.W., Burger, D., Wang, C.: Auto- Gen: Enabling next-gen LLM applications via multi-agent conversation. Preprint arXiv:2308.08155 [cs.AI].https://doi.org/10.48550/arXiv.2308.08155

    Pith/arXiv arXiv doi:10.48550/arxiv.2308.08155

  69. [69]

    ACM Transactions on Software Engineering and Methodology (2025).https://doi.or g/10.1145/3769676

    Xu, H., Wang, S., Li, N., Wang, K., Zhao, Y., Chen, K., Yu, T., Liu, Y., Wang, H.: Large language models for cyber security: A systematic literature review. ACM Transactions on Software Engineering and Methodology (2025).https://doi.or g/10.1145/3769676

    doi:10.1145/3769676 2025

  70. [70]

    Sensors25(6), 1666:1–1666:40 (2025).https://doi.org/10.3390/s2 5061666

    Yigit, Y., Ferrag, M., Ghanem, M.C., Sarker, I.H., Maglaras, L.A., Chrysoulas, C., Moradpoor, N., Tihanyi, N., Janicke, H.: Generative AI and LLMs for critical infrastructure protection: Evaluation benchmarks, agentic AI, challenges, and op- portunities. Sensors25(6), 1666:1–1666:40 (2025).https://doi.org/10.3390/s2 5061666

    work page doi:10.3390/s2 2025

  71. [71]

    In: Proceedings of the Network and Distributed System Security Symposium (NDSS)

    Yu, J., Xie, X., Hu, Q., Ma, Y., Zhao, Z.: Chimera: Harnessing multi-agent LLMs for automatic insider threat simulation. In: Proceedings of the Network and Distributed System Security Symposium (NDSS). Internet Society (2026). https://doi.org/10.14722/ndss.2026.240375

    arXiv 2026

  72. [72]

    IEEE Intelligent Systems40, 30–37 (2025).https://doi.org/10.1109/mis.2024.351 7792

    Zhang, Q., Liu, Z., Pan, S., Wang, C.: The rise of small language models. IEEE Intelligent Systems40, 30–37 (2025).https://doi.org/10.1109/mis.2024.351 7792

    work page doi:10.1109/mis.2024.351 2025

  73. [73]

    arXiv:2506.12508v5 [cs.AI] (2026),https:// arxiv.org/abs/2506.12508v5

    Zhang, W., Zeng, L., Xiao, Y., Li, Y., Cui, C., Zhao, Y., Hu, R., Liu, Y., Zhou, Y., An, B.: AgentOrchestra: Orchestrating multi-agent intelligence with the tool- environment-agent (TEA) protocol. arXiv:2506.12508v5 [cs.AI] (2026),https:// arxiv.org/abs/2506.12508v5

    Pith/arXiv arXiv 2026

  74. [74]

    Journal of King Saud University Computer and Information Sciences37(10), 330:1–330:19 (2025)

    Zhou, Y., Chen, Y.: Adaptive heterogeneous multi-agent debate for enhanced educational and factual reasoning in large language models. Journal of King Saud University Computer and Information Sciences37(10), 330:1–330:19 (2025). https://doi.org/10.1007/s44443-025-00353-3 32 A. ElZemity et al

    work page doi:10.1007/s44443-025-00353-3 2025

  75. [75]

    Journal of Universal Computer Science30(9), 1163–1183 (2024)

    Çetin, O., Ekmekcioglu, E., Arief, B., Hernandez-Castro, J.: An empirical eval- uation of large language models in static code analysis for PHP vulnerabil- ity detection. Journal of Universal Computer Science30(9), 1163–1183 (2024). https://doi.org/10.3897/jucs.134739

    work page doi:10.3897/jucs.134739 2024