It is shown that the quality of the generated risk indicators is the main driver of zero-shot cyber threat detection performance, and that retrieval mainly benefits weaker LLMs by generating more discriminative risk indicators, whereas stronger models achieve comparable performance without retrieved context.
Abstract
We propose a two-stage large language model (LLM) framework for zero-shot detection of insider threats and advanced persistent threats (APTs) from heterogeneous security logs. The framework models user activity as chronological timelines and incorporates retrieval-augmented generation (RAG) to provide personalised behavioural context from each user's historical activity. Rather than performing end-to-end classification directly from raw logs, it first generates structured, interpretable sets of threat-specific risk indicators, which are then classified jointly across temporal sequences to capture attack patterns spanning multiple windows.The framework is evaluated on two benchmark datasets, CERT r5.2 for insider threat detection and PicoDomain for APT detection, using four combinations of two open-weight LLMs under both retrieval and non-retrieval settings. All configurations outperform the previous state-of-the-art LLM-based framework (GABM), with the best configuration improving the F1-score by 11.40 percentage points on CERT r5.2 and 31.50 percentage points on PicoDomain. Results further show that retrieval mainly benefits weaker LLMs by generating more discriminative risk indicators, whereas stronger models achieve comparable performance without retrieved context. The most effective assignment of LLMs to the two stages depends on the dataset. These findings show that the quality of the generated risk indicators is the main driver of zero-shot cyber threat detection performance.
Experimental results show that LLMs, when guided by rubric-based prompts and supplemented with ATT&CK domain knowledge, achieve robust performance across detection, localization, and TTP mapping tasks.
Joon-Young Gwak, Aubrey Strier, Zhaohan Xi et al.· 0 citations
Security teams must prioritize vulnerabilities before exploitation evidence is complete. Existing signals, such as CVSS, EPSS, advisories, and public exploits, are useful but fragmented and time-sensitive; retrospective rankings can therefore overstate performance by using evidence unavailable at decision time. We present ThreatLens, a simple yet effective and deployment-realistic framework for CVE prioritization. ThreatLens ranks vulnerabilities at each review point using only cutoff-valid evidence and learns from future CISA KEV entries as weak supervision for exploitation relevance. Under forward-in-time, CVE-disjoint evaluation, ThreatLens significantly outperforms CVSS, EPSS, and rule-based evidence-fusion baselines. On the held-out test split, ThreatLens surfaces 80.0% of future KEV CVEs in the top 20, over three times EPSS at the same budget, and reaches 95.9% in the top 50. Early-warning analysis further shows that ThreatLens identifies a substantial fraction of subsequent KEV entries before formal catalog inclusion, supporting timely, evidence-grounded triage.
Soroush Motamedi Sedeh, Panteha Shahrivar, M. Qureshi et al.· 0 citations
: Insider threats remain difficult to detect because malicious actions often resemble legitimate user behaviour and may evolve gradually over time. This paper presents MLBRS, a multi-layer behavioural risk scoring framework that combines rule-based scoring, statistical deviation analysis, and Isolation Forest-based anomaly detection to generate continuous employee-level risk scores. The framework integrates behavioural indicators, personalised deviation modelling, and multivariate anomaly detection to identify both abrupt and gradual behavioural changes. Due to the limited availability of publicly accessible datasets containing database-query-level insider threat activity, a synthetic dataset was constructed to simulate organisational behaviour with temporal consistency, multiple employee roles, and diverse attack scenarios. Existing insider-threat datasets primarily capture system-level activity and do not adequately represent database interactions. Experimental evaluation demonstrates consistent detection performance, achieving an ROC-AUC of 0.978 and an F1-score of 0.88 on the synthetic dataset. Additional cross-dataset evaluation using CERT-derived behavioural traces shows reduced but stable performance under less aligned behavioural conditions. The results indicate that MLBRS provides an interpretable and scalable approach for behavioural insider threat detection across heterogeneous activity patterns.
V. L. Kartheek, Aayush Shah, Rishav Jain et al.· International Conference on...· 0 citations
An Intrusion Detection System (IDS)-contextual ensemble learning framework that assigns network traffic to three operationally meaningful risk tiers: High, Medium and Low is presented.
Reeta Mishra, Neelu Chaudhary· Journal of Trends in Compute...· 0 citations
Results show that DeepScrub improves fraud review accuracy, reduces first-stage review workload, and provides traceable evidence for production risk-review workflows, showing that domain adaptation can matter more than model scale in this setting.
Large language models (LLMs) embedded in enterprise workflows cannot structurally distinguish legitimate instructions from adversarial ones in the same token stream, making prompt injection OWASP's top LLM risk for two consecutive editions a persistent threat across direct and indirect vectors. This paper presents PromptShield-RT, a layered, real-time, model-agnostic framework combining input normalization and provenance tagging, lexical-heuristic pattern matching, a statistical classifier, structural anomaly features, and calibrated risk fusion, with policy-driven mitigation (allow/sanitize/quarantine/block) and an explainable, adaptive-feedback mechanism for SOC workflows. We construct an original evaluation corpus, SynPI-Bench (n = 450, six categories), and a template-disjoint held-out generalization set (n = 31) with novel phrasings, obfuscation encodings, and adversarial hard-negative benign text. Using template-grouped 5-fold cross-validation, the fused pipeline achieves 92.4% accuracy (F1 = 0.930, AUC = 0.990), outperforming heuristic-only (57.0%) and naive-averaged (59.2%) baselines, while a lexical classifier reaches 85.9% with lower precision. We report a pronounced generalization gap on the held-out set (48.4% accuracy, 90% false-positive rate on hard negatives), quantifying a known limitation of surface-lexical defenses. The pipeline achieves sub-millisecond P95 latency (0.266 ms), within typical 50 ms enterprise SLAs. We situate PromptShield-RT relative to structural, architectural, and guardrail-product defenses, arguing for layered, defense-in-depth architectures, with reproducible code provided.
Fatimah Alhamzawi· Al-Noor Journal of Engineeri...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
