Stabilising Explainability Fragility in Cybersecurity AI: The Impact and Mitigation of Multicollinearity in Public Benchmark Datasets
arXiv:2605.22529v1 Announce Type: new Abstract: This paper investigates a unexplored yet impactful vulnerability in AI explainability used in intrusion detection (IDS): multicollinearity-induced instability. Despite extensive reliance on post-hoc explainability tools such as SHAP or LIME, the impact of correlated features on explanation robustness is not evaluated. We introduce a formal theorem stating that multicollinearity inflates attribution variance. This demonstrates that explanations and feature importances are non-identifiable under multicollinearity. A suite of comprehensive experimen
This research addresses a critical and previously unexplored vulnerability in AI explainability, coinciding with the increasing deployment of AI in sensitive applications like cybersecurity.
The findings reveal fundamental instability in current AI explainability techniques under common data conditions, necessitating a re-evaluation of models and their interpretability in critical domains.
Trust in AI explanations, particularly in cybersecurity, is diminished, requiring new methods for robust explainability and potentially leading to a shift in how AI systems are designed and audited.
- · Explainable AI researchers
- · Cybersecurity AI developers focused on robust models
- · Regulators setting AI safety standards
- · Developers relying solely on current post-hoc explainability
- · Organizations deploying black-box AI in critical systems
- · Cybersecurity AI products with unaddressed explainability fragility
Immediate re-evaluation of current cybersecurity AI systems and their explainability components will occur.
Development of new, provably robust explainability methods will be accelerated, fostering a new standard for AI trustworthiness.
Regulatory bodies may impose stricter explainability requirements for AI used in high-stakes applications, potentially slowing AI adoption in some sectors until solutions are mature.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG