Explainable and Trustworthy Artificial Intelligence for Cybersecurity: A Framework for Transparent Malware Detection and Automated Threat Intelligence
Security teams increasingly rely on machine learning to detect malware and on language models and other automation to process threat intelligence. Both use cases pose the question, though: when should we trust what the system says? This paper proposes a set of trusting, explainable and trustworthy artificial intelligence in cybersecurity, including transparent detection of malware and evidence-based automated threat intelligence, that combine seven trust properties (fidelity, stability, robustness, calibration, plausibility, accountability, and security of explanations itself), compares eight families of explanation methods, shows an alert-level workflow where every explanation is verified before it impacts a decision, and describes an intelligence pipeline where every claim is tied to a source and to a technique in a standard taxonomy. It lists levels of automation for six intelligence tasks and an evaluation protocol for detection, explanations, calibration, analyst utility, extraction quality, and governance. The article describes the evidence that explanations may disagree and are unstable or manipulated, and that language model extraction of threat techniques may be highly error prone to error modes. It views explanations as an attack surface and a source of over-reliance, not just a benefit. It presents an analysis work in design and does not provide any new experimental results. Four tables and three figures show the framework concrete, and can be adopted by security operations.
Authors
- Sidhanta Panigrahy (ORCID: https://orcid.org/0009-0001-1044-8875)
- Gokul Selvam Sadasivam
Institutions
- University of California, Berkeley (US)
Publication Details
- Journal
- International Journal of Scientific Research in Science Engineering and Technology
- Published
- 2026-11-15
- DOI
- https://doi.org/10.32628/ijsrset2613606
- Primary Topic
- Explainable Artificial Intelligence (XAI)
- Type
- article
- Field-Weighted Citation Impact
- 0.00