Phishing Detection

TypeSafe AI's Jev is a new kind of safety judge. It generates no text and no label: it answers questions the developer defines in advance and returns a probability for each answer. The vendor admits that adversarial input can change these answers but gives no numbers. In this paper, we present the first controlled adversarial evaluation of Jev. We build a test harness that sends every system the same payload under the same policy, and run it on four public-data tasks, from rule-based ones (input moderation, injection detection, routing) to a judgment-based one (phishing detection), with fifty attack variants from thirteen families. The targets are Jev, five GPT and GLM judges, and two local classifiers (Llama Guard 3, Prompt Guard 2). We find that Jev is the most robust system on input moderation: eight forgeries of its structured input bypass none of its decisions, and a sentence asking only why someone might make a harmful request bypasses 9% of its blocks, against 56% for GPT-6 Astra. On phishing detection. However, the same forgeries bypass 19-29% of the emails it flags, and even without attack Jev detects only 62% of phishing emails, against 77-81% for the GLM judges. Jev is also the fastest judge, but it flags the most benign requests (27%). The local classifiers fail differently: a training-example statement makes Llama Guard 3 miss 62% of the phishing it had flagged. These results suggest that Jev resists attacks on tasks with a clear rule to check but is more exposed on tasks that call for an overall judgment.

Phishing remains one of the most common vectors for financial and identity fraud, and most detection systems still rely on inspecting a page's URL, HTML markup, or domain registration history. These signals are easy for an attacker to rotate or obfuscate, and they say very little about what actually convinces a victim to hand over a password or a card number: the way the page looks. This paper describes a visual, image-based approach to phishing detection that treats a rendered webpage the same way a human eye would, as a picture that either matches a trusted brand or doesn't. A YOLOv8 convolutional neural network was trained to classify full-page website screenshots as phishing or legitimate based on layout, logo placement, color scheme, and login-form structure, rather than on text extracted from the page. The system reached 92% classification accuracy on a held-out test set, processed a single screenshot in roughly 100 milliseconds, and, after a round of data augmentation aimed specifically at lighting, compression, and scaling variation, cut the false-positive rate by 11% relative to the pre-augmentation baseline. The paper walks through the dataset construction, the augmentation strategy, the model architecture and training setup, and the resulting performance, and closes with a discussion of where this kind of visual detector fits alongside, rather than instead of, existing URL- and content-based defenses.

Email spam and phishing attacks remain a critical security threat. Adversaries increasingly exploit large language models to craft contextually convincing malicious messages, and existing spam detection systems often struggle to keep pace. Generalization across diverse and evolving attack scenarios is limited, which reduces effectiveness once these systems are deployed in practice. This paper introduces Adaptive Uncertainty-Routed Analysis (AURA), a multimodal email threat detection system that analyzes both the content of an email and its embedded URLs. AURA is built around two layers: the first quantifies prediction uncertainty from a URL classifier, and only ambiguous messages are escalated to a fine-tuned transformer encoder for semantic analysis. The system is evaluated on eight heterogeneous training corpora together with two held-out real-world corpora spanning a decade of adversarial campaigns. AURA reaches a macro F1-score of 0.9858 in-distribution, and on NazPhish-Eval and GuenterTrap-Eval it maintains 0.9502 and 0.9436, respectively, which is evidence of robust generalization under genuine distribution shift.

Conventional in-distribution evaluation can overestimate robustness when training and test data share recurring task-specific patterns or surface cues. This risk is especially relevant in social-engineering fraud detection, where attackers can preserve malicious intent while changing the scenario, impersonated entity, or wording. We study this problem as scenario-level out-of-distribution (SL-OOD) detection for SMS and voice phishing, where entire attack scenarios are held out from training while the label space remains fixed. This setting tests whether models can generalize to unseen attack scenarios using decision-relevant evidence rather than familiar scenario-specific cues. Using this SL-OOD evaluation, we find that high in-distribution performance does not reliably predict held-out robustness across feature-, encoder-, and decoder-based baselines. We interpret this gap as scenario memorization: reliance on recurring scenario-specific lexical or entity cues rather than decision-relevant evidence. We propose ECoG, an evidence-consistent generative framework that combines evidence-span supervision with a rationale-label consistency objective during training. On the 0.5B decoder, relative to the same backbone trained without consistency regularization, ECoG raises Macro-F1 on OOD challenging instances by 3.22 points, reduces the share of predictions whose generated rationale supports the opposite label by 4.22 points, and increases token-level overlap with reference evidence spans by 8.38 points; the reduction in prediction-rationale inconsistency is consistent across four decoder backbones. These results suggest that compact generative detectors can benefit from evidence supervision and rationale-label consistency under social-engineering shift.

Phishing emails remain one of the most persistent cybersecurity threats, and machine-learning classifiers are widely used to detect them. Most reported detection accuracies, however, are measured on clean, in-distribution test data rather than on emails deliberately altered to evade detection. This paper reports a controlled, pairwise comparison of two phishing-detection approaches a TF-IDF + Logistic Regression baseline and a fine-tuned DistilBERT transformer trained on a unified corpus of 82,255 emails drawn from six public datasets and evaluated under three conditions: normal in-distribution, synthetic phishing, and adversarial phishing. Both models exceeded 98% accuracy on clean data yet degraded sharply under adversarial testing: TF-IDF + LR fell to 64.00% (a 34.59-percentage-point drop) and DistilBERT fell to 63.64% (a 35.40-percentage-point drop) a gap of only 0.36 percentage points, equivalent to a single email in the 275-sample adversarial test set. LIME, SHAP, and attention-rollout analysis indicate the two models relied on different evidence yet showed similar vulnerability. Pairwise error analysis shows the models agreed on 54.9% of adversarial samples but each made a similar number of exclusive errors (24 and 25 respectively), indicating partly complementary rather than identical failure modes. The results show that clean-data accuracy does not predict adversarial robustness, and that adversarial testing should be a standard part of phishing-detection evaluation.

Following the rapid progress of generative Artificial Intelligence, there is a growing threat posed by conversational scams. These scams often span over multiple weeks or months, gradually build trust and request for money or sensitive information. Existing scam-detection systems mainly focus on isolated messages, which renders them inadequate against this evolving threat. This paper extends single-message phishing detection and presents an explainable agentic system for detecting sophisticated conversational scams. It also introduces ConScamBench-278, an initial public multi-category benchmark for conversational scam detection spanning eight scam types, released to support reproducible evaluation and future expansion. On isolated messages the single-message detector attains 100% phishing recall, while the conversation-level detector identifies all conversational scams in the public LoveFraud02 corpus (83/83) and reaches 97.8% accuracy (95% CI [95.4, 99.0]) on ConScamBench-278. Two user studies (N = 100 and N = 45) further motivate the system: participants report frequently experiencing uncertainty when judging suspicious conversations. In an uncontrolled pre/post comparison, users self-reported trust, self-confidence, and perceived need for AI-based scam detection all increased (p < 0.001, Wilcoxon signed-rank). The system also receives a System Usability Scale score of 74.7 (95% CI [72.5, 76.9]), above the established usability benchmark.

Phishing is a multi-modal threat. We present a hybrid pipeline that scores each modality with its own engine and fuses the results. Three engines are built, deployed, and independently benchmarked: a four-stage URL stack (Domain Guard, lexical model, threat intelligence, and an asymmetric L2 fusion sidecar); a generalization-hardened DistilBERT NLP classifier whose held-out real-phishing recall rises from 0.8% to 87.3%; and a threat-intelligence synchronizer with end-to-end OpenTelemetry instrumentation confirming 1:1 message conservation. A decision-level fusion stage, characterized on a 10,677-email whole-system benchmark, reaches F1=0.914 with a calibrated probabilistic-OR over URL, header, and phishing-probability channels while reducing held-out real-spam false positives to 3.6%. Because that benchmark uses proxy URL and header channels and an operating point still needing recalibration, we present it as a preliminary integrated result. For deployable detection, the limiting factor is how well a model generalizes, not how accurately it scores data drawn from its own training distribution.

The expansion of the digital domain has resulted in a substantial increase in digital communication, with email emerging as one of the most prominent channels. The proliferation of email communication is apparent in both professional and personal contexts, thereby creating numerous vulnerabilities for malicious actors to exploit. Spam emails, a form of unsolicited correspondence often bearing malicious intent towards recipients, have been an ongoing challenge for email users since the inception of email technology, and this problem has been exacerbated by the growth of the digital landscape. Email spam filters are integral components of email clients, engineered to identify potentially harmful messages and alert users to their malicious content. Phishing, frequently the initial phase of malware-based attacks, is evolving rapidly, with malware becoming increasingly sophisticated over time. A widely adopted approach for detecting malicious activity within malware and spam domains is the application of machine learning. Our aim is to assess the impact of the evolution within the spam email domain on these machine learning-based detection systems and to explore strategies for mitigating associated performance degradation.

Phishing attacks remain a major cybersecurity threat, exploiting deceptive URLs to steal sensitive user information. Traditional blacklist and rule-based detection approaches are reactive and often fail to identify newly emerging phishing URLs. This paper proposes a lightweight hybrid framework for real-time phishing URL detection that combines blacklist-based screening with a Multi-Layer Perceptron (MLP) classifier operating solely on structural URL features. The framework extracts 16 URL-derived features capturing structural, domain-based, and security-related characteristics without requiring webpage content access, third-party APIs, or visual rendering, making it computationally efficient for real-time deployment. The system was trained and evaluated on the PhiUSIIL phishing dataset containing 235,795 labelled URLs. Experimental results show that the proposed MLP achieved 99.24% accuracy, 98.74% precision, 99.95% recall, 99.34% F1-score, and 99.65% ROC-AUC, outperforming Random Forest, Logistic Regression, XGBoost, LightGBM, and CatBoost under the same evaluation setting. The hybrid architecture achieved an average inference latency of 1.2 ms per URL and a peak throughput of 4,200 URLs per second under concurrent processing. A functional desktop application prototype, CyberGuard, further demonstrates deployment viability. The results indicate that the proposed framework provides an accurate and computationally efficient solution for real-time phishing URL detection in resource-constrained environments.

In phishing detection, machine learning classifiers act as a first line of defense, but the false positives they produce are triaged by human analysts. The excessive false alarms cause alert fatigue that erodes human oversight. We propose CoGReV, a hybrid framework that augments standard machine learning classifiers with a post-hoc non-monotonic reasoning layer implemented in Answer Set Programming. The layer applies a confidence-gated defeasible rule that revises a phishing prediction toward legitimate only when website metadata is present and the classifier's decision is low-confidence, deferring uncertain predictions to the reasoning layer while leaving out confident decisions to the classifiers. This gating acts as a function-allocation mechanism between the classifier and the reasoning layer. Unlike an ungated rule, which reduces false positives only by discarding genuine detections and degrades phishing recall by about nine percentage points, the proposed gated rule keeps recall within 0.7 percentage points of the no-revision baseline for all classifiers while revising only 0.27 percent of decisions. By lowering false alarms, without sacrificing detection, CoGReV reduces analysts alert fatigue and integrates new domain knowledge into the reasoning layer in $\mathcal{O}(n)$ time.

Organizations usually calibrate phishing detectors on public corpora, because privacy rules make their own labeled legitimate mail hard to collect. This practice breaks the false-positive budget of trained detectors: calibrated for a 1% false-positive rate (FPR) on older public mail and deployed on DataPhish 2025, they realize 20.2-22.9% FPR, 9-94 times the budget across three calibration sources. The cause is calibration transfer, not ranking: the classifiers separate the deployment mail well, but their thresholds do not transfer. We introduce CyberCane, a privacy-bounded neuro-symbolic detector that scores email only by similarity to known phishing in a redacted, training-only index, plus deterministic rules, and calibrates verdicts to an explicit FPR budget. Because legitimate mail never enters its score, only its threshold depends on the calibration source. Under the same shift CyberCane realizes 5.4% FPR at 86.4% recall, reducing the overrun 2.7-6.1 times across sources, seeds, a second shift family, and a local embedder. A segment-level variant, scoring an email's most phishing-like part, lowers the realized FPR a further 1.2-2.7 times at equal ranking quality and withstands benign padding. CyberCane screens an email locally in 0.20 s, sending no text out; a 70B zero-shot LLM ranks better at strict FPR but is 14-24 times slower and must receive each email. We also show that LLM rewriting does not evade embedding-based detectors and that a random split of DataPhish 2025 leaks 96% of its test set. All protocols were fixed in advance, and the released splits and code let a deployer measure realized FPR before trusting a budget.

Modern phishing campaigns increasingly evade snapshot-based URL classifiers using interaction gates (e.g., checkbox/slider challenges), delayed content rendering, and logo-less credential harvesters. This shifts URL triage from static classification toward an interactive forensics task: an analyst must actively navigate the page while isolating themselves from potential runtime exploits. We present TraceScope, a decoupled triage pipeline that operationalizes this workflow at scale. To prevent the observer effect and ensure safety, a sandboxed operator agent drives a real GUI browser guided by visual motivation to elicit page behavior, freezing the session into an immutable evidence bundle. Separately, an adjudicator agent circumvents LLM context limitations by querying evidence on demand to verify a MITRE ATT&CK checklist, and generates an audit-ready report with extracted indicators of compromise (IOCs) and a final verdict. Evaluated on 708 reachable URLs from existing dataset (241 verified phishing from PhishTank and 467 benign from Tranco-derived crawling), TraceScope achieves 0.94 precision and 0.78 recall, substantially improving recall over three prior visual/reference-based classifiers while producing reproducible, analyst-grade evidence suitable for review. More importantly, we manually curated a dataset of real-world phishing emails to evaluate our system in a practical setting. Our evaluation reveals that TraceScope demonstrates superior performance in a real-world scenario as well, successfully detecting sophisticated phishing attempts that current state-of-the-art defenses fail to identify.

System prompt configuration can make the difference between near-total phishing blindness and near-perfect detection in LLM email agents. We present PhishNChips, a study of 11 models under 10 prompt strategies, showing that prompt-model interaction is a first-order security variable: a single model's phishing bypass rate ranges from under 1% to 97% depending on how it is configured, while the false-positive cost of the same prompt varies sharply across models. We then show that optimizing prompts around highly predictive signals can improve benchmark performance, reaching up to 93.7% recall at 3.8% false positive rate, but also creates a brittle attack surface. In particular, domain-matching strategies perform well when legitimate emails mostly have matched sender and URL domains, yet degrade sharply when attackers invert that signal by registering matching infrastructure. Response-trace analysis shows that 98% of successful bypasses reason in ways consistent with the inverted signal: the models are following the instruction, but the instruction's core assumption has become false. A counter-intuitive corollary follows: making prompts more specific can degrade already-capable models by replacing broader multi-signal reasoning with exploitable single-signal dependence. We characterize the resulting tension between detection, usability, and adversarial robustness as a navigable tradeoff, introduce Safetility, a deployability-aware metric that penalizes false positives, and argue that closing the adversarial gap likely requires tool augmentation with external ground truth.

Phishing detectors built on engineered website features attain near-perfect accuracy under i.i.d.\ evaluation, yet deployment security depends on robustness to post-deployment feature manipulation. We study this gap through a cost-aware evasion framework that models discrete, monotone feature edits under explicit attacker budgets. Three diagnostics are introduced: minimal evasion cost (MEC), the evasion survival rate $S(B)$, and the robustness concentration index (RCI). On the UCI Phishing Websites benchmark (11,055 instances, 30 ternary features), Logistic Regression, Random Forests, Gradient Boosted Trees, and XGBoost all achieve $\mathrm{AUC}\ge 0.979$ under static evaluation. Under budgeted sanitization-style evasion, robustness converges across architectures: the median MEC equals 2 with full features, and over 80% of successful minimal-cost evasions concentrate on three low-cost surface features. Feature restriction improves robustness only when it removes all dominant low-cost transitions. Under strict cost schedules, infrastructure-leaning feature sets exhibit 17-19% infeasible mass for ensemble models, while the median MEC among evadable instances remains unchanged. We formalize this convergence: if a positive fraction of correctly detected phishing instances admit evasion through a single feature transition of minimal cost $c_{\min}$, no classifier can raise the corresponding MEC quantile above $c_{\min}$ without modifying the feature representation or cost model. Adversarial robustness in phishing detection is governed by feature economics rather than model complexity.

Phishing remains the most pervasive threat to the Web, enabling large-scale credential theft and financial fraud through deceptive webpages. While recent reference-based and generative-AI-driven phishing detectors achieve strong accuracy, their reliance on external knowledge bases, cloud services, and complex multimodal pipelines fundamentally limits practicality, scalability, and reproducibility. In contrast, conventional deep learning approaches often fail to generalize to evolving phishing campaigns. We introduce SpecularNet, a novel lightweight framework for reference-free web phishing detection that demonstrates how carefully designed compact architectures can rival heavyweight systems. SpecularNet operates solely on the domain name and HTML structure, modeling the Document Object Model (DOM) as a tree and leveraging a hierarchical graph autoencoding architecture with directional, level-wise message passing. This design captures higher-order structural invariants of phishing webpages while enabling fast, end-to-end inference on standard CPUs. Extensive evaluation against 13 state of the art phishing detectors, including leading reference-based systems, shows that SpecularNet achieves competitive detection performance with dramatically lower computational cost. On benchmark datasets, it reaches an F1 score of 93.9%, trailing the best reference-based method slightly while reducing inference time from several seconds to approximately 20 milliseconds per webpage. Field and robustness evaluations further validate SpecularNet in real-world deployments, on a newly collected 2026 open-world dataset, and against adversarial attacks.

Misclassifications in spam and phishing detection are very harmful, as false negatives expose users to attacks while false positives degrade trust. Existing uncertainty-based detectors can flag potential errors, but possibly be deceived and offer limited interpretability. This paper presents X-MAP, an eXplainable Misclassification Analysis and Profilling framework that reveals topic-level semantic patterns behind model failures. X-MAP combines SHAP-based feature attributions with non-negative matrix factorization to build interpretable topic profiles for reliably classified spam/phishing and legitimate messages, and measures each message's deviation from these profiles using Jensen-Shannon divergence. Experiments on SMS and phishing datasets show that misclassified messages exhibit at least two times larger divergence than correctly classified ones. As a detector, X-MAP achieves up to 0.98 AUROC and lowers the false-rejection rate at 95% TRR to 0.089 on positive predictions. When used as a repair layer on base detectors, it recovers up to 97% of falsely rejected correct predictions with moderate leakage. These results demonstrate X-MAP's effectiveness and interpretability for improving spam and phishing detection.