pith. sign in

arxiv: 2507.06252 · v2 · pith:2B47GHV7new · submitted 2025-07-05 · 💻 cs.CR · cs.AI· cs.LG

False Alarms, Real Damage: Adversarial Attacks Using LLM-based Models on Text-based Cyber Threat Intelligence Systems

Samaneh Shafee , Alysson Bessani , Pedro M. Ferreira This is my paper

Pith reviewed 2026-05-25 07:52 UTC · model grok-4.3

classification 💻 cs.CR cs.AIcs.LG
keywords adversarial attackscyber threat intelligenceCTI pipelineevasion attacksLLM-generated textfake contentmachine learning securityOSINT
0
0 comments X

The pith

Adversarial LLM-generated fake text can mislead classifiers throughout cyber threat intelligence pipelines.

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

The paper shows that adversarial text generation can produce fake cybersecurity content capable of evading detection in full CTI systems that collect and analyze open-source data. This leads to misclassification of indicators, degraded selection accuracy, and functional disruption across the pipeline. The work emphasizes evasion attacks because they open the door to flooding and poisoning. A reader would care since these automated systems depend on unvetted external text feeds that can be manipulated at scale.

Core claim

Adversarial text generation techniques can create fake cybersecurity and cybersecurity-like text that misleads classifiers, degrades performance, and disrupts system functionality. The focus is primarily on the evasion attack, as it precedes and enables flooding and poisoning attacks within the CTI pipeline.

What carries the argument

The evasion attack using LLM-based adversarial text generation applied to the full CTI pipeline that ingests open-source textual inputs.

If this is right

  • Evasion attacks degrade the information selection capabilities of CTI systems.
  • Generated fake text causes classifiers to select incorrect indicators of compromise.
  • System functionality is disrupted when adversarial content enters the pipeline.
  • Flooding and poisoning attacks become feasible once evasion succeeds.

Where Pith is reading between the lines

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

  • CTI operators may need to add text-origin checks or generation detectors to their ingestion stage.
  • The same generation techniques could affect other open-source intelligence systems beyond cybersecurity.
  • Empirical tests on live CTI tools with controlled fake inputs would quantify the scale of performance loss.

Load-bearing premise

CTI pipelines ingest textual inputs from open sources that may include fake or manipulated content and lack built-in protections against such manipulation.

What would settle it

Run an experiment feeding LLM-generated fake CTI reports into an existing classifier and measure the resulting increase in false IoC detections or drop in accuracy compared to real reports.

Figures

Figures reproduced from arXiv: 2507.06252 by Alysson Bessani, Pedro M. Ferreira, Samaneh Shafee.

Figure 1
Figure 1. Figure 1: Taxonomy of input text in a CTI Pipeline. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Proposed integrated CTI extraction pipeline. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Illustration of generating adversarial text using the attention mechanism and ChatGPT-4o. The generated adversarial text is input to a pre [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Flooding attack workflow. 5.6 Poisoning attack In the proposed CTI pipeline ( [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Distribution of semantic similarity scores within the dataset of [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Kernel Density Estimation Plot of real tweet and FaN text gradi [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗
read the original abstract

Cyber Threat Intelligence (CTI) has emerged as a vital complementary approach that operates in the early phases of the cyber threat lifecycle. CTI involves collecting, processing, and analyzing threat data to provide a more accurate and rapid understanding of cyber threats. Due to the large volume of data, automation through Machine Learning (ML) and Natural Language Processing (NLP) models is essential for effective CTI extraction. These automated systems leverage Open Source Intelligence (OSINT) from sources like social networks, forums, and blogs to identify Indicators of Compromise (IoCs). Although prior research has focused on adversarial attacks on specific ML models, this study expands the scope by investigating vulnerabilities within various components of the entire CTI pipeline and their susceptibility to adversarial attacks. These vulnerabilities arise because they ingest textual inputs from various open sources, including real and potentially fake content. We analyse three types of attacks against CTI pipelines, including evasion, flooding, and poisoning, and assess their impact on the system's information selection capabilities. Specifically, on fake text generation, the work demonstrates how adversarial text generation techniques can create fake cybersecurity and cybersecurity-like text that misleads classifiers, degrades performance, and disrupts system functionality. The focus is primarily on the evasion attack, as it precedes and enables flooding and poisoning attacks within the CTI pipeline.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The manuscript investigates vulnerabilities in text-based Cyber Threat Intelligence (CTI) pipelines to adversarial attacks generated by LLM-based models. It examines three attack types—evasion, flooding, and poisoning—with primary emphasis on evasion attacks that produce fake cybersecurity or cybersecurity-like text. The central claim is that such attacks can mislead classifiers, degrade performance, and disrupt system functionality because CTI systems ingest textual inputs from open sources including potentially fake content; evasion is positioned as an enabler for the other attacks.

Significance. If the empirical results on attack success rates, performance degradation, and pipeline disruption are robustly demonstrated with appropriate controls and metrics, the work would be significant for the cybersecurity community. It would provide concrete evidence of risks in automated OSINT-based CTI extraction and could motivate development of defenses such as input sanitization or adversarial training for threat intelligence classifiers.

major comments (2)
  1. [Abstract] Abstract: the claims that adversarial text 'misleads classifiers, degrades performance, and disrupts system functionality' and that evasion 'precedes and enables flooding and poisoning' are stated at a high level without any methods, datasets, quantitative results, or error analysis, so the support for the central empirical claims cannot be evaluated.
  2. [Methodology (inferred from structure)] No section provides the specific LLM-based adversarial text generation techniques, the definition of the CTI pipeline components under test, the classifiers or NLP models targeted, or the evaluation metrics (e.g., precision, recall, or detection rates), all of which are load-bearing for validating the reported impacts.
minor comments (1)
  1. [Abstract] The abstract could more explicitly distinguish the scope of 'cybersecurity and cybersecurity-like text' with a brief example to clarify the threat model.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful review and for highlighting the need for greater specificity. We agree that the current abstract and methodology presentation are too high-level to allow full evaluation of the empirical claims, and we will revise the manuscript to address this.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claims that adversarial text 'misleads classifiers, degrades performance, and disrupts system functionality' and that evasion 'precedes and enables flooding and poisoning' are stated at a high level without any methods, datasets, quantitative results, or error analysis, so the support for the central empirical claims cannot be evaluated.

    Authors: We agree that the abstract, as written, presents the central claims at too high a level. In the revised version we will add a concise sentence referencing the specific LLM generation approach, the datasets used for evaluation, and the key quantitative outcomes (attack success rates and performance degradation) while remaining within abstract length limits. revision: yes

  2. Referee: [Methodology (inferred from structure)] No section provides the specific LLM-based adversarial text generation techniques, the definition of the CTI pipeline components under test, the classifiers or NLP models targeted, or the evaluation metrics (e.g., precision, recall, or detection rates), all of which are load-bearing for validating the reported impacts.

    Authors: We accept the referee's observation that the current manuscript does not supply these load-bearing details. We will add an explicit 'Experimental Setup' subsection that defines the LLM prompting techniques for adversarial text generation, decomposes the CTI pipeline into its ingestion/processing/analysis stages, names the targeted NLP classifiers, and lists the evaluation metrics (precision, recall, F1, attack success rate, and pipeline disruption measures). revision: yes

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper is an empirical study of adversarial attacks on CTI pipelines using LLM-generated text. It contains no mathematical derivations, fitted parameters presented as predictions, self-citations used as load-bearing uniqueness theorems, or ansatzes smuggled via prior work. The central claims rest on experimental demonstrations of evasion, flooding, and poisoning attacks rather than any chain that reduces to its own inputs by construction. The reader's assessment of score 1.0 aligns with the absence of any load-bearing circular steps.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The abstract introduces no free parameters, no new axioms, and no invented entities; it describes an empirical investigation of existing attack surfaces in CTI systems.

pith-pipeline@v0.9.0 · 5778 in / 1186 out tokens · 40894 ms · 2026-05-25T07:52:15.459440+00:00 · methodology

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

66 extracted references · 66 canonical work pages · 3 internal anchors

  1. [1]

    What are the attackers doing now? automating cyberthreat intelligence extraction from text on pace with the changing threat landscape: A survey,

    M. R. Rahman, R. M. Hezaveh, and L. Williams, “What are the attackers doing now? automating cyberthreat intelligence extraction from text on pace with the changing threat landscape: A survey,” ACM Computing Surveys, vol. 55, no. 12, pp. 1–36, 2023

    work page 2023

  2. [2]

    Exploring emerging hacker assets and key hackers for proactive cyber threat intelligence,

    S. Samtani, R. Chinn, H. Chen, and J. F. Nunamaker Jr, “Exploring emerging hacker assets and key hackers for proactive cyber threat intelligence,” Journal of Manage- ment Information Systems , vol. 34, no. 4, pp. 1023–1053, 2017. 13

    work page 2017

  3. [3]

    # twiti: Social listening for threat intel- ligence,

    H. Shin, W. Shim, S. Kim, S. Lee, Y. G. Kang, and Y. H. Hwang, “# twiti: Social listening for threat intel- ligence,” in Proceedings of the Web Conference 2021, 2021, pp. 92–104

    work page 2021

  4. [4]

    A novel approach for detection and ranking of trendy and emerging cyber threat events in twitter streams,

    A. Bose, V . Behzadan, C. Aguirre, and W. H. Hsu, “A novel approach for detection and ranking of trendy and emerging cyber threat events in twitter streams,” in Proceedings of the 2019 IEEE/ACM International Confer- ence on Advances in Social Networks Analysis and Mining , 2019, pp. 871–878

    work page 2019

  5. [5]

    Follow the blue bird: A study on threat data published on twitter,

    F. Alves, A. Andongabo, I. Gashi, P . M. Ferreira, and A. Bessani, “Follow the blue bird: A study on threat data published on twitter,” in Computer Security– ESORICS 2020: 25th European Symposium on Research in Computer Security, ESORICS 2020, Guildford, UK, September 14–18, 2020, Proceedings, Part I 25 . Springer, 2020, pp. 217–236

    work page 2020

  6. [6]

    Timiner: Automatically extracting and analyzing cat- egorized cyber threat intelligence from social data,

    J. Zhao, Q. Yan, J. Li, M. Shao, Z. He, and B. Li, “Timiner: Automatically extracting and analyzing cat- egorized cyber threat intelligence from social data,” Comput. Secur., vol. 95, p. 101867, 2020

    work page 2020

  7. [7]

    A comparative study on cyber threat intelligence: The security incident response perspective,

    D. Schlette, M. Caselli, and G. Pernul, “A comparative study on cyber threat intelligence: The security incident response perspective,” IEEE Communications Surveys & Tutorials, vol. 23, no. 4, pp. 2525–2556, 2021

    work page 2021

  8. [8]

    Machine-generated text: A comprehensive survey of threat models and detection methods,

    E. N. Crothers, N. Japkowicz, and H. L. Viktor, “Machine-generated text: A comprehensive survey of threat models and detection methods,” IEEE Access , vol. 11, pp. 70 977–71 002, 2023

    work page 2023

  9. [9]

    A new text classification model based on contrastive word embed- ding for detecting cybersecurity intelligence in twitter,

    H.-S. Shin, H.-Y. Kwon, and S.-J. Ryu, “A new text classification model based on contrastive word embed- ding for detecting cybersecurity intelligence in twitter,” Electronics, vol. 9, no. 9, p. 1527, 2020

    work page 2020

  10. [10]

    Looking beyond iocs: Automatically extracting attack patterns from external cti,

    M. T. Alam, D. Bhusal, Y. Park, and N. Rastogi, “Looking beyond iocs: Automatically extracting attack patterns from external cti,” in Proceedings of the 26th International Symposium on Research in Attacks, Intrusions and Defenses, 2023, pp. 92–108

    work page 2023

  11. [11]

    Vulcan: Automatic extraction and analysis of cyber threat intelligence from unstruc- tured text,

    H. Jo, Y. Lee, and S. Shin, “Vulcan: Automatic extraction and analysis of cyber threat intelligence from unstruc- tured text,” Computers & Security , vol. 120, p. 102763, 2022

    work page 2022

  12. [12]

    Ex- tractor: Extracting attack behavior from threat reports,

    K. Satvat, R. Gjomemo, and V . Venkatakrishnan, “Ex- tractor: Extracting attack behavior from threat reports,” in 2021 IEEE European Symposium on Security and Pri- vacy (EuroS&P). IEEE, 2021, pp. 598–615

    work page 2021

  13. [13]

    Cyber threat intelligence for soc analysts,

    N. Rastogi and M. T. Alam, “Cyber threat intelligence for soc analysts,” 2023

    work page 2023

  14. [14]

    Towards end-to-end cyberthreat detection from twit- ter using multi-task learning,

    N. Dion ´ısio, F. Alves, P . M. Ferreira, and A. Bessani, “Towards end-to-end cyberthreat detection from twit- ter using multi-task learning,” in 2020 international joint conference on neural networks (IJCNN) . IEEE, 2020, pp. 1–8

    work page 2020

  15. [15]

    Processing tweets for cybersecurity threat awareness,

    F. Alves, A. Bettini, P . M. Ferreira, and A. Bessani, “Processing tweets for cybersecurity threat awareness,” Information Systems, vol. 95, p. 101586, 2021

    work page 2021

  16. [16]

    A machine learning-based fintech cyber threat attribu- tion framework using high-level indicators of compro- mise,

    U. Noor, Z. Anwar, T. Amjad, and K.-K. R. Choo, “A machine learning-based fintech cyber threat attribu- tion framework using high-level indicators of compro- mise,” Future Generation Computer Systems , vol. 96, pp. 227–242, 2019

    work page 2019

  17. [17]

    Collect- ing indicators of compromise from unstructured text of cybersecurity articles using neural-based sequence labelling,

    Z. Long, L. Tan, S. Zhou, C. He, and X. Liu, “Collect- ing indicators of compromise from unstructured text of cybersecurity articles using neural-based sequence labelling,” in 2019 international joint conference on neural networks (IJCNN). IEEE, 2019, pp. 1–8

    work page 2019

  18. [18]

    Cyber threat intelligence modeling based on heterogeneous graph convolutional network,

    J. Zhao, Q. Yan, X. Liu, B. Li, and G. Zuo, “Cyber threat intelligence modeling based on heterogeneous graph convolutional network,” in 23rd international symposium on research in attacks, intrusions and defenses (RAID 2020), 2020, pp. 241–256

    work page 2020

  19. [19]

    Atdg: An automatic cy- ber threat intelligence extraction model of dpcnn and bigru combined with attention mechanism,

    B. Cui, J. Li, and W. Hou, “Atdg: An automatic cy- ber threat intelligence extraction model of dpcnn and bigru combined with attention mechanism,” in Interna- tional Conference on Web Information Systems Engineering. Springer, 2023, pp. 189–204

    work page 2023

  20. [20]

    S. EMK. (2020) CTI extractor – ECHO network. [Online]. Available: https://www.echocti.com/en/

    work page 2020

  21. [21]

    From logs to stories: Human-centred data mining for cyber threat intelligence,

    N. Afzaliseresht, Y. Miao, S. Michalska, Q. Liu, and H. Wang, “From logs to stories: Human-centred data mining for cyber threat intelligence,” IEEE Access , vol. 8, pp. 19 089–19 099, 2020

    work page 2020

  22. [22]

    From threat reports to continuous threat intelligence: a comparison of attack technique extraction methods from textual artifacts,

    M. R. Rahman and L. Williams, “From threat reports to continuous threat intelligence: a comparison of attack technique extraction methods from textual artifacts,” arXiv preprint arXiv:2210.02601, 2022

    work page arXiv 2022

  23. [23]

    To- wards an automated dissemination process of cyber threat intelligence data using stix,

    O. Briliyant, N. P . Tirsa, and M. A. Hasditama, “To- wards an automated dissemination process of cyber threat intelligence data using stix,” 2021 6th Interna- tional Workshop on Big Data and Information Security (IWBIS), pp. 109–114, 2021

    work page 2021

  24. [24]

    Ctibench: A benchmark for evaluating llms in cy- ber threat intelligence,

    M. T. Alam, D. Bhushl, L. Nguyen, and N. Rastogi, “Ctibench: A benchmark for evaluating llms in cy- ber threat intelligence,” arXiv preprint arXiv:2406.07599, 2024

    work page arXiv 2024

  25. [25]

    Evidence-based prioritization of cybersecurity threats,

    R. Kerkdijk, S. Tesink, F. Fransen, and F. Fal- conieri, “Evidence-based prioritization of cybersecurity threats,” 2021

    work page 2021

  26. [26]

    Alert prioritisation in security operations centres: A systematic survey on criteria and methods,

    F. Jalalvand, M. Baruwal Chhetri, S. Nepal, and C. Paris, “Alert prioritisation in security operations centres: A systematic survey on criteria and methods,” ACM Computing Surveys, 2024

    work page 2024

  27. [27]

    Se- curity operations center: A systematic study and open challenges,

    M. Vielberth, F. B ¨ohm, I. Fichtinger, and G. Pernul, “Se- curity operations center: A systematic study and open challenges,” Ieee Access , vol. 8, pp. 227 756–227 779, 2020

    work page 2020

  28. [28]

    Creating cybersecurity knowledge graphs from mal- ware after action reports,

    A. Piplai, S. Mittal, A. Joshi, T. Finin, J. Holt, and R. Zak, “Creating cybersecurity knowledge graphs from mal- ware after action reports,” IEEE Access , vol. 8, pp. 211 691–211 703, 2020

    work page 2020

  29. [29]

    Mining threat intel- ligence about open-source projects and libraries from code repository issues and bug reports,

    L. Neil, S. Mittal, and A. Joshi, “Mining threat intel- ligence about open-source projects and libraries from code repository issues and bug reports,” in 2018 IEEE International Conference on Intelligence and Security Infor- matics (ISI). IEEE, 2018, pp. 7–12

    work page 2018

  30. [30]

    Azse- cure hacker assets portal: Cyber threat intelligence and malware analysis,

    S. Samtani, K. Chinn, C. Larson, and H. Chen, “Azse- cure hacker assets portal: Cyber threat intelligence and malware analysis,” in 2016 IEEE conference on intelli- gence and security informatics (ISI) . Ieee, 2016, pp. 19– 24

    work page 2016

  31. [31]

    (2020) Cyber threat intelligence (CTI): Analysis, dissemination, and feedback

    zvelo. (2020) Cyber threat intelligence (CTI): Analysis, dissemination, and feedback. [Online]. Available: https: 14 //zvelo.com/cti-analysis-dissemination-feedback/

    work page 2020

  32. [32]

    Advanced persistent threat group correlation analysis via attack behavior patterns and rough sets,

    J. Li, J. Liu, and R. Zhang, “Advanced persistent threat group correlation analysis via attack behavior patterns and rough sets,” Electronics, vol. 13, no. 6, p. 1106, 2024

    work page 2024

  33. [33]

    A taxonomy and survey of attacks against machine learning,

    N. Pitropakis, E. Panaousis, T. Giannetsos, E. Anas- tasiadis, and G. Loukas, “A taxonomy and survey of attacks against machine learning,” Computer Science Review, vol. 34, p. 100199, 2019

    work page 2019

  34. [34]

    Adversarial deep ensemble: Evasion attacks and defenses for malware detection,

    D. Li and Q. Li, “Adversarial deep ensemble: Evasion attacks and defenses for malware detection,” IEEE Transactions on Information Forensics and Security, vol. 15, pp. 3886–3900, 2020

    work page 2020

  35. [35]

    A survey for restricting the ddos traffic flooding and worm attacks in internet,

    R. Saranya, S. S. Kannan, and N. Prathap, “A survey for restricting the ddos traffic flooding and worm attacks in internet,” in 2015 International Conference on Applied and Theoretical Computing and Communication Technology (iCATccT), 2015, pp. 251–256

    work page 2015

  36. [36]

    Machine learning security against data poisoning: Are we there yet?

    A. E. Cin `a, K. Grosse, A. Demontis, B. Biggio, F. Roli, and M. Pelillo, “Machine learning security against data poisoning: Are we there yet?” Computer, vol. 57, no. 3, pp. 26–34, 2024

    work page 2024

  37. [37]

    A comprehensive survey on poisoning attacks and countermeasures in machine learning,

    Z. Tian, L. Cui, J. Liang, and S. Yu, “A comprehensive survey on poisoning attacks and countermeasures in machine learning,” ACM Computing Surveys , vol. 55, no. 8, pp. 1–35, 2022

    work page 2022

  38. [38]

    A survey of black-box adversarial attacks on computer vision models,

    S. Bhambri, S. Muku, A. Tulasi, and A. B. Buduru, “A survey of black-box adversarial attacks on computer vision models,” arXiv preprint arXiv:1912.01667, 2019

    work page arXiv 1912

  39. [39]

    Cyberthreat detection from twitter using deep neural networks,

    N. Dion ´ısio, F. Alves, P . M. Ferreira, and A. Bessani, “Cyberthreat detection from twitter using deep neural networks,” in 2019 international joint conference on neural networks (IJCNN). IEEE, 2019, pp. 1–8

    work page 2019

  40. [40]

    A survey on predictions of cyber-attacks utilizing real-time twitter tracing recog- nition,

    S. Altalhi and A. Gutub, “A survey on predictions of cyber-attacks utilizing real-time twitter tracing recog- nition,” Journal of Ambient Intelligence and Humanized Computing, pp. 1–13, 2021

    work page 2021

  41. [41]

    Evaluation of llm-based chatbots for osint-based cyber threat aware- ness,

    S. Shafee, A. Bessani, and P . M. Ferreira, “Evaluation of llm-based chatbots for osint-based cyber threat aware- ness,” Expert Systems with Applications, p. 125509, 2024

    work page 2024

  42. [42]

    Generating fake cyber threat intelligence using transformer-based models,

    P . Ranade, A. Piplai, S. Mittal, A. Joshi, and T. Finin, “Generating fake cyber threat intelligence using transformer-based models,” in 2021 International Joint Conference on Neural Networks (IJCNN) . IEEE, 2021, pp. 1–9

    work page 2021

  43. [43]

    Argh! automated rumor generation hub,

    L. Huynh, T. Nguyen, J. Goh, H. Kim, and J. B. Hong, “Argh! automated rumor generation hub,” in Proceedings of the 30th ACM International Conference on Information & Knowledge Management , 2021, pp. 3847– 3856

    work page 2021

  44. [44]

    Defending against neural fake news,

    R. Zellers, A. Holtzman, H. Rashkin, Y. Bisk, A. Farhadi, F. Roesner, and Y. Choi, “Defending against neural fake news,” Advances in neural information pro- cessing systems, vol. 32, 2019

    work page 2019

  45. [45]

    Generating natural language adversarial ex- amples on a large scale with generative models,

    Y. Ren, J. Lin, S. Tang, J. Zhou, S. Yang, Y. Qi, and X. Ren, “Generating natural language adversarial ex- amples on a large scale with generative models,” in ECAI 2020. IOS Press, 2020, pp. 2156–2163

    work page 2020

  46. [46]

    The Llama 3 Herd of Models

    A. Dubey, A. Jauhri, A. Pandey, A. Kadian, A. Al-Dahle, A. Letman, A. Mathur, A. Schelten, A. Yang, A. Fan et al. , “The llama 3 herd of models,” arXiv preprint arXiv:2407.21783, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  47. [47]

    Increas- ing diversity while maintaining accuracy: Text data generation with large language models and human interventions,

    J. J. Y. Chung, E. Kamar, and S. Amershi, “Increas- ing diversity while maintaining accuracy: Text data generation with large language models and human interventions,” arXiv preprint arXiv:2306.04140, 2023

    work page arXiv 2023

  48. [48]

    Utilizing prompt engineering to operationalize cybersecurity,

    K. Huang, G. Huang, Y. Duan, and J. Hyun, “Utilizing prompt engineering to operationalize cybersecurity,” in Generative AI Security: Theories and Practices . Springer, 2024, pp. 271–303

    work page 2024

  49. [49]

    Secure- bert: A domain-specific language model for cybersecu- rity,

    E. Aghaei, X. Niu, W. Shadid, and E. Al-Shaer, “Secure- bert: A domain-specific language model for cybersecu- rity,” in International Conference on Security and Privacy in Communication Systems. Springer, 2022, pp. 39–56

    work page 2022

  50. [50]

    Cyberpal. ai: Empowering llms with expert- driven cybersecurity instructions,

    M. Levi, Y. Alluouche, D. Ohayon, and A. Puzanov, “Cyberpal. ai: Empowering llms with expert- driven cybersecurity instructions,” arXiv preprint arXiv:2408.09304, 2024

    work page arXiv 2024

  51. [51]

    Long short-term memory,

    A. Graves and A. Graves, “Long short-term memory,” Supervised sequence labelling with recurrent neural net- works, pp. 37–45, 2012

    work page 2012

  52. [52]

    Ifnd: a benchmark dataset for fake news detection,

    D. K. Sharma and S. Garg, “Ifnd: a benchmark dataset for fake news detection,” Complex & intelligent systems , vol. 9, no. 3, pp. 2843–2863, 2023

    work page 2023

  53. [53]

    Do perceptually aligned gradients imply robustness?

    R. Ganz, B. Kawar, and M. Elad, “Do perceptually aligned gradients imply robustness?” in International Conference on Machine Learning . PMLR, 2023, pp. 10 628–10 648

    work page 2023

  54. [54]

    Robust kernel density estima- tion,

    J. Kim and C. D. Scott, “Robust kernel density estima- tion,” The Journal of Machine Learning Research , vol. 13, no. 1, pp. 2529–2565, 2012

    work page 2012

  55. [55]

    The wasserstein distance and approx- imation theorems,

    L. R ¨uschendorf, “The wasserstein distance and approx- imation theorems,” Probability Theory and Related Fields, vol. 70, no. 1, pp. 117–129, 1985

    work page 1985

  56. [56]

    Statistical aspects of wasserstein distances,

    V . M. Panaretos and Y. Zemel, “Statistical aspects of wasserstein distances,” Annual review of statistics and its application, vol. 6, no. 1, pp. 405–431, 2019

    work page 2019

  57. [57]

    Longformer: The Long-Document Transformer

    I. Beltagy, M. E. Peters, and A. Cohan, “Longformer: The long-document transformer,” arXiv preprint arXiv:2004.05150, 2020

    work page internal anchor Pith review Pith/arXiv arXiv 2004

  58. [58]

    Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context

    Z. Dai, Z. Yang, Y. Yang, J. Carbonell, Q. V . Le, and R. Salakhutdinov, “Transformer-xl: Attentive language models beyond a fixed-length context,” arXiv preprint arXiv:1901.02860, 2019

    work page internal anchor Pith review Pith/arXiv arXiv 1901

  59. [59]

    Kgv: Integrating large language models with knowledge graphs for cyber threat intelligence credibility assessment,

    Z. Wu, F. Tang, M. Zhao, and Y. Li, “Kgv: Integrating large language models with knowledge graphs for cyber threat intelligence credibility assessment,” arXiv preprint arXiv:2408.08088, 2024

    work page arXiv 2024

  60. [60]

    Triple-r: Automatic reasoning for fact verification using language models,

    M. Kanaani, “Triple-r: Automatic reasoning for fact verification using language models,” in Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC- COLING 2024), 2024, pp. 16 831–16 840

    work page 2024

  61. [61]

    (2025) WHOIS domain lookup - find website owners - GoDaddy IE

    GoDaddy team. (2025) WHOIS domain lookup - find website owners - GoDaddy IE. [Online]. Available: https://www.godaddy.com/en/offers/whois-b

    work page 2025

  62. [62]

    (2025) WHOIS search, domain name, website, and IP tools - who.is

    Who.is. (2025) WHOIS search, domain name, website, and IP tools - who.is. [Online]. Available: https: //who.is/

    work page 2025

  63. [63]

    Argh! automated rumor generation hub,

    L. Huynh, T. Nguyen, J. Goh, H. Kim, and J. B. Hong, “Argh! automated rumor generation hub,” in Proceedings of the 30th ACM International Conference on Information & Knowledge Management , ser. CIKM 15 ’21. New York, NY, USA: Association for Computing Machinery, 2021, p. 3847–3856. [Online]. Available: https://doi.org/10.1145/3459637.3481894

    work page doi:10.1145/3459637.3481894 2021

  64. [64]

    {EaTVul}:{ChatGPT-based} evasion attack against software vulnerability detection,

    S. Liu, D. Cao, J. Kim, T. Abraham, P . Mon- tague, S. Camtepe, J. Zhang, and Y. Xiang, “{EaTVul}:{ChatGPT-based} evasion attack against software vulnerability detection,” in 33rd USENIX Se- curity Symposium (USENIX Security 24), 2024, pp. 7357– 7374

    work page 2024

  65. [65]

    Textjuggler: fooling text classification tasks by generating high-quality adver- sarial examples,

    H. Peng, Z. Wang, C. Wei, D. Zhao, G. Xu, J. Han, S. Guo, M. Zhong, and S. Ji, “Textjuggler: fooling text classification tasks by generating high-quality adver- sarial examples,” Knowledge-Based Systems, vol. 300, p. 112188, 2024

    work page 2024

  66. [66]

    Textguise: Adaptive adversarial example attacks on text classifi- cation model,

    G. Chang, H. Gao, Z. Yao, and H. Xiong, “Textguise: Adaptive adversarial example attacks on text classifi- cation model,” Neurocomputing, vol. 529, pp. 190–203, 2023. 16 Fig. A1. First prompt: Optimized final prompt to send ChatGPT -4o and its response. Second prompt: Testing ChatGpt as a classifier. APPENDIX A The prompt, shown in Figure A1, is carefully...

    work page 2023