Pith. sign in

REVIEW 3 major objections 6 minor 70 references

"Explain, Don't Just Warn!" -- A Real-Time Framework for Generating Phishing Warnings with Contextual Cues

T0 review · 3 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read This paper claims that replacing generic anti-phishing warnings with short, specific explanations of why a site is suspicious trains users to recognize phishing on their own—with the effect largest for novices and the warning engine…

desk verdict PXP is a genuinely useful system and the user study shows a real effect, but the 'even without any warning' headline claim overreaches because the transfer test is near-transfer on the same cue types. read the letter →

arxiv 2505.06836 v2 pith:2SLLS2KS submitted 2025-05-11 cs.CR

Sayak Saha Roy , Cesar Torres , Shirin Nilizadeh This is my paper · ORCID
classification cs.CR
keywords phishingwarningsexplainablesecuritycontextuallargelanguagemodelsuserstudydetectionmentalbrowser
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

PhishXplain claims that swapping a browser's generic 'this site is deceptive' warning for a short explanation of exactly which visible cues make the site suspicious trains users to recognize phishing on their own. In a controlled study, participants who saw these explainable warnings correctly classified 6.44 of 8 unassisted websites on average, versus 5.16 for those who saw generic warnings, with the gap driven by phishing sites and by the least security-savvy users. The same warning generator ran for a month on 7,091 live phishing sites, producing warnings for 94% and correctly describing features in about 96% of sampled indicators. If the transfer effect is real, explainable warnings turn a moment of blocking into a micro-lesson, and do it locally on consumer hardware without sending browsing data to a third party.

What carries the argument

PhishXplain's engine is a two-step prompt pipeline around a fixed lookup table of 26 user-visible phishing cues. A small local language model first reads the fully rendered page source, with HTML elements wrapped in unique delimiters, and returns up to four matching cues as a JSON object; a second prompt then fills pre-written template sentences with concrete artifacts (typos, URLs, demands) from the flagged code. A separate renderer re-draws the page with colored bounding boxes on those elements and captures a screenshot, so the warning shows both text and visual proof. The lookup table is what keeps a small model grounded and reproducible: the model only maps to known features and fills in blanks, rather than generating open-ended explanations.

What would settle it

Run the same two-group study but assemble the unassisted phishing set from cues absent from the warning phase—for example QR-code payment lures or brand impersonation through lookalike HTTPS subdomains with no obvious URL anomaly. If Group A no longer outperforms Group B on those sites, the claimed transfer is cue-specific practice, not a generalizable detection skill.

Watch

Extended reading notes

Core claim

The paper's central claim is that contextual, feature-specific warnings—not just a generic block page—cause users to build or activate phishing cues they can apply later, without any warning. The evidence is a between-subjects study in which Group A saw PhishXplain warnings that named, explained, and visually highlighted up to four suspicious elements, while Group B saw a standard generic block-page warning; both groups then judged eight unassisted websites. Group A averaged 6.44/8 correct versus 5.16 for Group B (p < 0.01 for phishing sites), and low-proficiency users jumped from 5.0 to 6.8 correct, nearly matching high-proficiency users. Group A also named more correct cues in unassisted phishing sites (7.84 versus 5.45). The paper does not claim the tool improves detection itself; it claims the explanation transforms user responses, and that this can be done in about 5 seconds with 2.7GB of memory on a mid-range consumer machine.

Load-bearing premise

The strongest result depends on the unassisted phishing sites sharing the same cue types the explainable warnings had just practiced; if those shared cues are removed, the learning advantage may vanish.

Editorial extensions

If this is right

  • If the result holds, anti-phishing tools can convert every block into a brief teaching moment without a separate training program.
  • Low-proficiency users gain the most, so explainable warnings could narrow the security-skill gap that generic warnings leave untouched.
  • The 94% warning-generation rate and 96% feature-correctness indicate most blocklisted phishing pages contain visible, explainable cues; only backend-only evasions are unexplainable.
  • Because false-positive rates on benign sites did not improve, the training effect is specific to recognizing malicious cues rather than general caution.
  • The local, 5-second, roughly 2.7GB deployment makes the approach practical as a privacy-preserving browser extension rather than a cloud API.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A likely mechanism is cue rehearsal rather than durable mental-model formation: the four unassisted phishing sites shared cue types with the warning phase, so a delayed re-test with novel cues is needed to tell the two apart.
  • Adding explicit 'looks legitimate because...' signals for benign sites would be a natural extension; the paper's own limitations note that explaining only malicious cues leaves benign judgments unchanged.
  • The same two-step lookup-table scaffolding could be reused for other warnings (malware, credential theft, tech-support scams), with the table swapped for the target domain's cue inventory.
  • If future phishing campaigns shift to cues outside the 26-entry table, coverage is testable: the system's 94% generation rate and 96% correctness are only as current as the table.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. This paper presents PhishXplain (PXP), a browser-extension framework that augments existing anti-phishing tools by replacing generic warnings with contextual explanations produced by a local 4-bit quantized LLaMA 3.2:3B model. PXP parses the rendered DOM, uses a two-prompt structured architecture constrained by a 26-feature lookup table, and generates annotated screenshots with up to four highlighted suspicious features. The authors evaluate latency, memory, and reliability on 100 live PhishTank sites, report a one-month longitudinal deployment over 7,091 GSB-flagged URLs with a 94% warning-generation rate and a 96% manually assessed correctness rate on a 500-site sample, and run a between-subjects user study (n=150) comparing PXP warnings with generic Safe Browsing warnings. The study reports that PXP participants were more accurate on unguided phishing websites (H1, p<0.01), reported higher confidence and helpfulness, and improved cue articulation, especially for low-proficiency users. The paper argues that contextual warnings support learning and transfer to unassisted detection.

Significance. The paper's direction is valuable: replacing blocklist warnings with concrete, user-visible cues is a plausible way to improve both immediate compliance and phishing cue awareness, and the local-LLM design with constrained output is a practical engineering contribution. The H1 phishing-only difference (p<0.01), the significant proficiency-by-warning interaction, and the release of the extension and codebook are strengths. However, the headline correctness figure is self-referential, the transfer test reuses the same cue categories, and one reported significant result (H2) does not meet the paper's own Bonferroni threshold. The practical contribution remains defensible, but the stronger generalizability claims in the abstract and conclusion are not yet supported.

major comments (3)
  1. [§4.2 and §4.3] The 96% correctness figure in the abstract and Section 4.3 rests on a circular design: the same two author-coders who constructed the 26-feature lookup table and template text (Section 3.3) manually reviewed the 500-site longitudinal sample, and the LLM is constrained by the prompt architecture to emit only features from that table. No inter-rater reliability is reported for the longitudinal sample, and no independent ground-truth source is used. As written, the 1,316/1,368 agreement measures consistency with the authors' own feature taxonomy rather than external correctness. Please report independent coding, inter-rater statistics, and disagreement resolution, or re-scope the correctness claim accordingly.
  2. [§5.2 and Table 2] The transfer evidence for the abstract's claim that users achieve higher accuracy "even without any warning" is near-transfer. The text states that the four unguided phishing websites "were chosen to contain malicious features similar to those highlighted in the warning phase (for Group A)," and Table 2 confirms complete cue overlap: Phishing-1 reuses IDN homograph and third-party hosting from Warning-2/Warning-3, Phishing-2 reuses suspicious URL and urgency from Warning-1, and the other test sites similarly reuse warning-phase cue categories. No unguided site introduces a cue category absent from the warning phase, so Group A's advantage is consistent with practiced feature recognition rather than a generalizable mental model. The paper should either add test sites with novel cue categories or restrict the claim to immediate transfer on previously encountered cue types.
  3. [§5.3, RQ2/H2] In RQ2 the authors report H2 as significant with p<0.015, but the Bonferroni-corrected significance threshold stated earlier in Section 5.3 for performance-related hypotheses (H1–H4) is α=0.0125. Since 0.015 > 0.0125, the reported p-value does not meet the stated threshold. Exact p-values and a corrected conclusion for H2 are needed, because this result currently appears in the abstract's confidence and trust claims.
minor comments (6)
  1. [§4.2] The reliability formula is typeset ambiguously as "Reliability = 10× CFR+(1−FMR)+AA+CSA /4" without parentheses; please display the intended expression, presumably 10 × (CFR + (1−FMR) + AA + CSA)/4.
  2. [§5.3, Table 5] Table 5 mislabels hypotheses relative to the text in Section 5.1: it pairs H3 with "Accuracy×Proficiency interaction," H4 with "Recognition of phishing cues," and H6 with "Perceived warning helpfulness," whereas the text defines H3 as perceived helpfulness, H4 as the low-proficiency interaction, H6 as cue recognition, and H5 as the helpfulness-by-proficiency interaction. Please correct the table to match the hypothesis definitions.
  3. [§5.3, RQ1] The sentence "90.67% of Group A participants correctly identified Phishing-1, compared to only 52% for Phishing-3" appears to compare different websites; the intended contrast is likely Group B's 52% on Phishing-1, so the wording should be corrected.
  4. [§5.3] Exact p-values and effect sizes should be reported for the Mann-Whitney and ANOVA results instead of inequalities such as p<0.01 and p<0.005, which makes it difficult to verify results against the stated Bonferroni thresholds.
  5. [§3.4] Section 3.4 references "Figure 7" twice for apparently different illustrations; the second reference should point to a distinct appendix figure or be renumbered.
  6. [§6] The limitations section acknowledges the absence of a long-term field study and the benign-site false-positive issue, but it should also address the circularity of the reliability assessment and the near-transfer design of the unguided test set, since these directly bound the headline claims.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the transfer-test design and shared coding taxonomy are generalizability limitations, not derivation steps that reduce to the paper's inputs.

full rationale

PhishXplain's derivation chain is empirical rather than self-referential. The 26-feature lookup table is built from a manually coded sample of 1,000 PhishTank sites (Section 3.3), and the LLM is constrained to output features from that table; however, the reported 94% warning-generation rate and 96% indicator correctness (Section 4.3) are measured against manual assessment of whether each emitted indicator is 'correctly aligned with the phishing characteristics of the websites' — not automatic consequences of the constraint. Reliability is computed via CFR, FMR, AA, and CSA against manually annotated ground truth (Section 4.2), so the coders can and do reject hallucinated features, as shown in the worked example in the Appendix. The user study's central result is an observed between-group difference on unguided classification (Group A 6.44 vs Group B 5.16 correct of eight; p<0.01); this could have been null, so it is not forced by construction. The two genuine validity concerns — Section 5.2's statement that the unguided phishing sites 'were chosen to contain malicious features similar to those highlighted in the warning phase (for Group A),' and Section 5.3's coding of open-ended justifications 'using features from our lookup table' — are limitations on generalization and possible cue-specific learning, not circular derivations: they do not make the measured outcome equal to the paper's inputs. The paper's own Limitations section (Section 6) explicitly restricts the user study to immediate impact and acknowledges the benign-site imbalance, which further supports treating the transfer result as near-transfer rather than as a forced derivation. The only overlapping-author citation ([54], used to justify the ten parsed HTML tags in Section 3.2) is a design motivation and is not load-bearing for the central user-learning claim. No step satisfies the standard of Eq. X = Eq. Y by construction or a fitted parameter renamed as a prediction.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The central contribution is a software system, so the ledger is mostly design choices rather than fitted physical constants. The main fitted parameters are the four-feature cap, the proficiency cutoffs, and the equal-weight reliability formula, each chosen by the authors and each influencing the reported numbers. The domain assumptions are about the completeness of the 26-feature taxonomy and the fidelity of DOM and screenshot capture. No new entities are postulated.

free parameters (4)
  • Maximum features per warning = 4
    Set so that 998 of 1,000 sampled PhishTank sites in the authors' own coding exercise had at most four features; this caps output and directly affects CFR/FMR and thus the reliability score.
  • Low/high proficiency cutoffs = <29 low, >36 high (out of 40)
    These thresholds partition participants for the H4/H5 subgroup analyses and were set by the authors following Schoni et al.; different cutoffs would change the interaction results.
  • Reliability score weights = Equal weights on CFR, 1-FMR, AA, CSA
    The composite reliability score is an unweighted average of four component metrics chosen by the authors; it determines the model ranking in Table 1.
  • Network idle delay before DOM capture = 500 ms
    Hand-selected wait after window.onload to capture late-loading content; affects which source code is sent to the LLM.
assumptions (4)
  • domain assumption All user-facing phishing indicators can be represented by the 26-category lookup table and the ten HTML tag types chosen by the authors.
    Sections 3.2-3.3: the parser extracts only p, ol, h*, a, iframe, ul, form, button, li, input, and the LLM is restricted to the 26 features; phishing cues outside this set are invisible to the system.
  • domain assumption Google Safe Browsing's blocklist flag is a sufficient trigger and the explained site has already been correctly identified as phishing.
    Section 3: PXP activates only after the default anti-phishing tool flags a page; its evaluation inherits GSB's false-positive and false-negative behavior.
  • domain assumption The fully rendered DOM captured in the browser, including post-onload content, accurately reflects the page the user sees.
    Section 3.1: the source-code capture uses window.onload plus 500 ms network idle; dynamic or backend-gated content that never appears in the DOM cannot be explained.
  • domain assumption A screenshot of a re-rendered page with injected CSS bounding boxes is a faithful depiction of the original page for warning purposes.
    Section 3.4: the annotated screenshot is captured after re-rendering with inserted highlighting; rendering differences could alter what the user sees.

how reviews work

0 comments
Cite this review

Pith. "Pith review of "Explain, Don't Just Warn!" -- A Real-Time Framework for Generating Phishing Warnings with Contextual Cues." pith.science (2026). https://pith.science/paper/2SLLS2KS

@misc{pith2026250506836,
  author       = {Pith},
  title        = {Pith review of: "Explain, Don't Just Warn!" -- A Real-Time Framework for Generating Phishing Warnings with Contextual Cues},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2SLLS2KS}},
  note         = {Machine review of arXiv:2505.06836}
}
read the original abstract

Anti-phishing tools typically display generic warnings that offer users limited explanation on why a website is considered malicious, which can prevent end-users from developing the mental models needed to recognize phishing cues on their own. This becomes especially problematic when these tools inevitably fail - particularly against evasive threats, and users are found to be ill-equipped to identify and avoid them independently. To address these limitations, we present PhishXplain (PXP), a real-time explainable phishing warning system designed to augment existing detection mechanisms. PXP empowers users by clearly articulating why a site is flagged as malicious, highlighting suspicious elements using a memory-efficient implementation of LLaMA 3.2. It utilizes a structured two-step prompt architecture to identify phishing features, generate contextual explanations, and render annotated screenshots that visually reinforce the warning. Longitudinally implementing PhishXplain over a month on 7,091 live phishing websites, we found that it can generate warnings for 94% of the sites, with a correctness of 96%. We also evaluated PhishXplain through a user study with 150 participants split into two groups: one received conventional, generic warnings, while the other interacted with PXP's explainable alerts. Participants who received the explainable warnings not only demonstrated a significantly better understanding of phishing indicators but also achieved higher accuracy in identifying phishing threats, even without any warning. Moreover, they reported greater satisfaction and trust in the warnings themselves. These improvements were especially pronounced among users with lower initial levels of cybersecurity proficiency and awareness. To encourage the adoption of this framework, we release PhishXplain as a browser extension.

Figures

Figures reproduced from arXiv: 2505.06836 by the authors.

Figure 1
Figure 1. The generic phishing warning shown by Google [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The PhishXplain framework 3 FRAMEWORK [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 4
Figure 4. A <p> tag encapsulated by an identifier. Note the URL is appended at the start as ELEMENT 0. The URL is visible since it is a placeholder URL. 3.3 Feature Extraction The parsed source-code is then passed to the LLM through a pair of structured prompts designed to extract information about mali￾cious features present in the website’s source. Most evasive phishing attacks are engineered to bypass anti-phishing crawler… view at source ↗
Figures from the paper (6 more)
Figure 3
Figure 3. Figure 3: Example of PhishXplain’s explainable blocklisting warning page showcasing contextual details about suspicious [PITH_FULL_IMAGE:figures/full_fig_p004_3.png]
Figure 5
Figure 5. Figure 5: The first prompt to extract features from the source [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: The second prompt to insert artifacts into the [PITH_FULL_IMAGE:figures/full_fig_p005_6.png]
Figure 7
Figure 7. Figure 7: The original website renrendered with the malicious [PITH_FULL_IMAGE:figures/full_fig_p005_7.png]
Figure 8
Figure 8. Figure 8: Warning-2 [PITH_FULL_IMAGE:figures/full_fig_p015_8.png]
Figure 9
Figure 9. Figure 9: Warning-3 Warning-4 [PITH_FULL_IMAGE:figures/full_fig_p015_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

70 extracted references · 59 canonical work pages

  1. [1]

    Bitdefender TrafficLight

    2020. Bitdefender TrafficLight. https://www.bitdefender.com/solutions/ trafficlight.html

    work page 2020

  2. [2]

    Google Safebrowsing

    2020. Google Safebrowsing. https://safebrowsing.google.com/

    work page 2020

  3. [3]

    PhishTank

    2020. PhishTank. https://www.phishtank.com/faq.php

    work page 2020

  4. [4]

    Avast Online Security & Privacy

    2024. Avast Online Security & Privacy. https://www.avast.com/en-us/avast- online-security#mac

    work page 2024

  5. [5]

    Hossein Abroshan, Jan Devos, Geert Poels, and Eric Laermans. 2021. Phishing Happens Beyond Technology: The Effects of Human Behaviors and Demographics on Each Step of a Phishing Process.IEEE Access 9 (2021), 44928–44949

    work page 2021

  6. [6]

    Meta AI. 2024. Llama 3.2 3B: A Multilingual Large Language Model. https://ollama.com/library/llama3.2:3b Accessed: 2025-04-07

    work page 2024

  7. [7]

    Devdatta Akhawe and Adrienne Porter Felt. 2013. Alice in warningland: A large-scale field study of browser security warning effectiveness. In22nd USENIX Security Symposium (USENIX Security 13). 257–272

    work page 2013

  8. [8]

    Mahmoud Alsharnouby, Fadi Alaca, and Sonia Chiasson. 2015. Why phishing still works: User strategies for combating phishing attacks. InProceedings of the 33rd An- nual ACM Conference on Human Factors in Computing Systems. ACM, 1237–1246

    work page 2015

Show all 70 references
  1. [9]

    Clara Ament. 2017. The ubiquitous security expert: Overconfidence in information security. (2017)

    2017

  2. [10]

    Joseph Aneke, Carmelo Ardito, and Giuseppe Desolda. 2021. Help the User Recognize a Phishing Scam: Design of Explanation Messages in Warning Interfaces for Phishing Attacks. InInternational Conference on Human-Computer Interaction. Springer, 403–416

    2021

  3. [11]

    Nalin AS Arachchilage and Steve Love. 2014. Security awareness of computer users: A phishing threat avoidance perspective.Computers in Human Behavior 38 (2014), 304–312

    2014

  4. [12]

    Mehdi Babagoli, Mohammad Pourmahmood Aghababa, and Vahid Solouk. 2019. Heuristic nonlinear regression strategy for detecting phishing websites. Soft Computing 23, 12 (2019), 4315–4327

    2019

  5. [13]

    Casey Inez Canfield, Baruch Fischhoff, and Alex Davis. 2016. Quantifying phishing susceptibility for detection and behavior decisions.Human factors 58, 8 (2016), 1158–1172

    2016

  6. [14]

    Rui Chen, Joana Gaia, and H Raghav Rao. 2020. An examination of the effect of recent phishing encounters on phishing susceptibility.Decision Support Systems 133 (2020), 113287

    2020

  7. [15]

    Giuseppe Desolda, Joseph Aneke, Carmelo Ardito, Rosa Lanzilotti, and Maria Francesca Costabile. 2023. Explanations in warning dialogs to help users defend against phishing attacks. International Journal of Human-Computer Studies 176 (2023), 103056

    2023

  8. [16]

    Selenium Python Bindings 2 documentation. [n. d.]. Selenium with Python — Selenium Python Bindings 2 documentation. https://selenium- python.readthedocs.io/

  9. [17]

    Serge Egelman, Lorrie Faith Cranor, and Jason Hong. 2008. You’ve been warned: an empirical study of the effectiveness of web browser phishing warnings. InProceed- ings of the SIGCHI Conference on Human Factors in Computing Systems. 1065–1074

    2008

  10. [18]

    Robert Friel and Atindriyo Sanyal. 2023. Chainpoll: A high efficacy method for llm hallucination detection. arXiv preprint arXiv:2310.18344 (2023)

    2023 arXiv

  11. [19]

    GlobalStats. [n. d.]. Browser Market Share Worldwide. https: //gs.statcounter.com/browser-market-share

  12. [20]

    Francesco Greco, Giuseppe Desolda, Andrea Esposito, et al. 2023. Explaining Phish- ing Attacks: An XAI Approach to Enhance User Awareness and Trust.. InITASEC

    2023

  13. [21]

    Reza Hassanpour, Erdogan Dogdu, Roya Choupani, Onur Goker, and Nazli Nazli

  14. [22]

    Fredrik Heiding, Bruce Schneier, Arun Vishwanath, Jeremy Bernstein, and Peter S Park. 2023. Devising and detecting phishing: Large language models vs. smaller human models. arXiv preprint arXiv:2308.12287 (2023)

    2023 arXiv

  15. [23]

    Explain, Don’t Just Warn!

    Grant Ho, Ariana Mirian, Elisa Luo, Khang Tong, Euyhyun Lee, Lin Liu, Christopher A Longhurst, Christian Dameff, Stefan Savage, and Geoffrey M Voelker. 2024. Understanding the Efficacy of Phishing Training in Practice. In 2025 IEEE Symposium on Security and Privacy (SP). IEEE ...

    2024

  16. [24]

    Lei Huang, Weijiang Yu, Weitao Ma, Weihong Zhong, Zhangyin Feng, Haotian Wang, Qianglong Chen, Weihua Peng, Xiaocheng Feng, Bing Qin, et al. 2025. A sur- vey on hallucination in large language models: Principles, taxonomy, challenges, and open questions. ACM Transactions on In...

    2025

  17. [25]

    Ollama Inc. 2025. Ollama: Get up and running with large language models. https://ollama.com/

    2025

  18. [26]

    Ganesh Jawahar, Muhammad Abdul-Mageed, Laks VS Lakshmanan, and Dujian Ding. 2023. Llm performance predictors are good initializers for architecture search. arXiv preprint arXiv:2310.16712 (2023)

    2023 arXiv

  19. [27]

    S Carolin Jeeva and Elijah Blessing Rajsingh. 2016. Intelligent phishing url detection using association rule mining. Human-centric Computing and Information Sciences 6, 1 (2016), 1–19

    2016

  20. [28]

    Renren Jin, Jiangcun Du, Wuwei Huang, Wei Liu, Jian Luan, Bin Wang, and Deyi Xiong. 2024. A comprehensive evaluation of quantization strategies for large language models. In Findings of the Association for Computational Linguistics ACL

    2024

  21. [29]

    Marianne Junger, Lidia Montoya, and Freya-Jade Overink. 2017. Priming and warnings are not effective to prevent social engineering attacks.Computers in Human Behavior 66 (2017), 75–87

    2017

  22. [30]

    Iacovos Kirlappos and M Angela Sasse. 2011. Security education against phishing: A modest proposal for a major rethink.IEEE Security & Privacy 10, 2 (2011), 24–32

    2011

  23. [31]

    Jan Kocoń, Igor Cichecki, Oliwier Kaszyca, Mateusz Kochanek, Dominika Szydło, Joanna Baran, Julita Bielaniewicz, Marcin Gruza, Arkadiusz Janz, Kamil Kanclerz, et al. 2023. ChatGPT: Jack of all trades, master of none. Information Fusion 99 (2023), 101861

    2023

  24. [32]

    Takashi Koide, Naoki Fukushi, Hiroki Nakano, and Daiki Chiba. 2023. Detecting phishing sites using chatgpt. arXiv preprint arXiv:2306.05816 (2023)

    2023 arXiv

  25. [33]

    Bin Liang, Miaoqiang Su, Wei You, Wenchang Shi, and Gang Yang. 2016. Cracking classifiers for evasion: A case study on the google’s phishing pages filter. In Proceedings of the 25th International Conference on World Wide Web. 345–356

    2016

  26. [34]

    Yun Lin, Ruofan Liu, Dinil Mon Divakaran, Jun Yang Ng, Qing Zhou Chan, Yiwen Lu, Yuxuan Si, Fan Zhang, and Jin Song Dong. 2021. Phishpedia: A Hybrid Deep Learning Based Approach to Visually Identify Phishing Webpages.. InUSENIX Security Symposium. 3793–3810

    2021

  27. [35]

    Ruofan Liu, Yun Lin, Xianglin Yang, Siang Hwee Ng, Dinil Mon Divakaran, and Jin Song Dong. 2022. Inferring Phishing Intention via Webpage Appearance and Dynamics: A Deep Vision Based Approach. In 30th{USENIX} Security Symposium ({USENIX} Security 21)

    2022

  28. [36]

    Scott M Lundberg and Su-In Lee. 2017. A unified approach to interpreting model predictions. Advances in neural information processing systems 30 (2017)

    2017

  29. [37]

    Khalid Mahmood. 2016. Do people overestimate their information literacy skills? A systematic review of empirical evidence on the Dunning-Kruger effect. Communications in Information Literacy 10, 2 (2016), 3

    2016

  30. [38]

    Nathan Malkin, Arunesh Mathur, Marian Harbach, and Serge Egelman. 2017. Personalized security messaging: Nudges for compliance with browser warnings. In 2nd european workshop on usable security. internet society

    2017

  31. [39]

    Ariana Martino, Michael Iannelli, and Coleen Truong. 2023. Knowledge injection to counter large language model (LLM) hallucination. InEuropean Semantic Web Conference. Springer, 182–185

    2023

  32. [40]

    John McAlaney and Peter J Hills. 2020. Understanding phishing email processing and perceived trustworthiness through eye tracking.Frontiers in Psychology 11 (2020), 1756

    2020

  33. [41]

    Microsoft. 2023. What is the Windows Subsystem for Linux? https://learn.microsoft.com/en-us/windows/wsl/about

    2023

  34. [42]

    Rami M Mohammad, Fadi Thabtah, and Lee McCluskey. 2014. Intelligent rule-based phishing websites classification. IET Information Security 8, 3 (2014), 153–160

    2014

  35. [43]

    Luong Anh Tuan Nguyen, Ba Lam To, Huu Khuong Nguyen, and Minh Hoang Nguyen. 2013. Detecting phishing web sites: A heuristic URL-based approach. In 2013 International Conference on Advanced Technologies for Communications (ATC 2013). IEEE, 597–602

    2013

  36. [44]

    Adam Oest, Kurt Thomas, Yi Wang, Adam Doupé, Gail-Joon Ahn, and Ziming Zhang. 2020. PhishTime: Continuous longitudinal measurement of the effectiveness of anti-phishing blacklists. Proceedings of the IEEE Symposium on Security and Privacy (S&P) (2020)

    2020

  37. [45]

    Adam Oest, Penghui Zhang, Brad Wardman, Eric Nunes, Jakub Burgis, Ali Zand, Kurt Thomas, Adam Doupé, and Gail-Joon Ahn. 2020. Sunrise to sunset: Analyzing the end-to-end life cycle and effectiveness of phishing attacks at scale. In 29th USENIX Security Symposium (USENIX Security 20)

    2020

  38. [46]

    OpenAI. 2021. Introducing the ChatGPT and Whisper APIs. https://openai.com/ blog/introducing-chatgpt-and-whisper-apis. [Accessed on 9 March 2023]

    2021

  39. [47]

    Kathryn Parsons, Dragana Calic, Malcolm Pattinson, Marcus Butavicius, Agata McCormac, and Tara Zwaans. 2017. The human aspects of information security questionnaire (HAIS-Q): two further validation studies. Computers & Security 66 (2017), 40–51

    2017

  40. [48]

    Justin Petelka, Yixin Zou, and Florian Schaub. 2019. Put your warning where your link is: Improving and evaluating email phishing warnings. InProceedings of the 2019 CHI conference on human factors in computing systems. 1–15

    2019

  41. [49]

    Rana Pourmohamad, Steven Wirsz, Adam Oest, Tiffany Bao, Yan Shoshitaishvili, Ruoyu Wang, Adam Doupé, and Rida A Bazzi. 2024. Deep Dive into Client-Side Anti-Phishing: A Longitudinal Study Bridging Academia and Industry. In Proceedings of the 19th ACM Asia Conference on Compute...

    2024

  42. [50]

    Prolific Researcher Help Center. 2023. What is informed consent? https://researcher-help.prolific.com/en/article/2273bd

    2023

  43. [51]

    Petri Puhakainen and Mikko Siponen. 2010. Improving employees’ compliance through information systems security training: an action research study. MIS quarterly (2010), 757–778

    2010

  44. [52]

    Routhu Srinivasa Rao and Alwyn Roshan Pais. 2019. Detection of phishing websites using an efficient feature-based machine learning framework.Neural Computing and Applications 31, 8 (2019), 3851–3873

    2019

  45. [53]

    Why should i trust you?

    Marco Tulio Ribeiro, Sameer Singh, and Carlos Guestrin. 2016. " Why should i trust you?" Explaining the predictions of any classifier. InProceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining. 1135–1144

    2016

  46. [54]

    Sayak Saha Roy and Shirin Nilizadeh. 2024. PhishLang: A lightweight, client-side phishing detection framework using MobileBERT for real-time, explainable threat mitigation. arXiv preprint arXiv:2408.05667 (2024)

    2024 arXiv

  47. [55]

    Sayak Saha Roy, Poojitha Thota, Krishna Vamsi Naragam, and Shirin Nilizadeh

  48. [56]

    Ozgur Koray Sahingoz, Ebubekir Buber, Onder Demir, and Banu Diri. 2019. Machine learning based phishing detection from URLs. Expert Systems with Applications 117 (2019), 345–357

    2019

  49. [57]

    Lorin Schöni, Victor Carles, Martin Strohmeier, Peter Mayer, and Verena Zimmermann. 2024. You Know What?-Evaluation of a Personalised Phishing Training Based on Users’ Phishing Knowledge and Detection Skills. InProceedings of the 2024 European Symposium on Usable Security. 1–14

    2024

  50. [58]

    Sakib Shahriar Shafin. 2024. An explainable feature selection framework for web phishing detection with machine learning. Data Science and Management (2024)

    2024

  51. [59]

    Kavya Sharma, Xinhui Zhan, Fiona Fui-Hoon Nah, Keng Siau, and Maggie X Cheng. 2021. Impact of digital nudging on information security behavior: an experimental study on framing and priming in cybersecurity. Organizational Cybersecurity Journal: Practice, Process and People 1, ...

    2021

  52. [60]

    PassMark Software. 2024. PassMark Software - PC Benchmark Charts. https://www.pcbenchmarks.net/

    2024

  53. [61]

    Lukas Tuggener, Pascal Sager, Yassine Taoudi-Benchekroun, Benjamin F Grewe, and Thilo Stadelmann. 2024. So you want your private LLM at home? A survey and benchmark of methods for efficient GPTs. In2024 11th IEEE Swiss Conference on Data Science (SDS). IEEE, 205–212

    2024

  54. [62]

    Igor Vayansky and Dan M Kumar. 2018. Phishing—Challenges and solutions. Computer Fraud & Security 2018, 1 (2018), 15–20

    2018

  55. [63]

    Norton Safe Web. [n. d.]. https://safeweb.norton.com/

  56. [64]

    Weining Yang, Jing Chen, Aiping Xiong, Robert W Proctor, and Ninghui Li. 2015. Effectiveness of a phishing warning in field settings. InProceedings of the 2015 Symposium and Bootcamp on the Science of Security. 1–2

    2015

  57. [65]

    Ying Yuan, Qingying Hao, Giovanni Apruzzese, Mauro Conti, and Gang Wang

  58. [66]

    Olga Zielinska, Allaire Welk, Christopher B Mayhorn, and Emerson Murphy-Hill

  59. [68]

    Are Adversarial Phishing Webpages a Threat in Reality?

    " Are Adversarial Phishing Webpages a Threat in Reality?" Understanding the Users’ Perception of Adversarial Webpages. InProceedings of the ACM on Web Conference 2024. 1712–1723

    2024

  60. [2015]

    As a valued customer, we’re giving you a special dis- count! -90% on all our offers, click here to view more!

    Exploring expert and novice mental models of phishing. In Proceedings of the 2015 Symposium and Bootcamp on the Science of Security. 1–2. SURVEY QUESTIONNAIRE Cybersecurity Habits (1) Please indicate your level of agreement with the following statements (Strongly Disagree to S...

    2015

  61. [2018]

    InProceedings of the ACMSE 2018 Conference

    Phishing e-mail detection by using deep learning algorithms. InProceedings of the ACMSE 2018 Conference. 1–1

    2018

  62. [2024]

    In2024 IEEE Symposium on Security and Privacy (SP)

    From chatbots to phishbots?: Phishing scam generation in commercial large language models. In2024 IEEE Symposium on Security and Privacy (SP). IEEE, 36–54

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.