AWARE-RG: Abstraction Aware Explanation of Resiliency Graphs for Cyber Physical Systems
CPS security analysts operate under need-to-know constraints that give them incomplete cross-domain views, risking flawed risk assessments. AWARE-RG addresses this by building a role-bounded abstraction lattice over Resiliency Graphs using MITRE ATT&CK tactics ordered by the ICS Cyber Kill Chain. It uses AI planning to validate attack-fault queries, repair incomplete analyst paths, and generate need-to-know explanations without exposing unauthorized predicates.
The paper presents a formally elegant framework that structurally enforces need-to-know disclosure for CPS attack-fault explanations through planning-based abstraction. However, its practical impact is constrained by an extremely narrow evaluation on a single synthetic scenario with author-crafted queries and no practitioner study, and by a reliance on LLM-based mapping that is only kept accurate via mandatory human expert review.
The formal properties are sound: Property 1 structurally guarantees that bridge nodes never expose opposite-domain $S_i$ predicates, and Property 2 confines every returned artifact to the analyst's authorized abstraction node. The lattice pruning is effective, reducing the HELM search space from $2^{90}$ to 31 nodes while preserving zero lexical leakage, and the inclusion of an adversarial threat model against the explanation interface itself ($T_1$--$T_3$) is a welcome security-aware touch. The core abstraction soundness claim correctly restates Sreedharan et al.'s HELM result.
The evaluation lacks external validity: it uses a single CPS (an LNG flare system), only 100 hand-designed queries, and the three authors themselves as the ground-truth experts, claiming 100% inter-rater agreement on node selection. This raises clear confirmation-bias concerns and offers no evidence that actual IT/OT analysts would find the explanations useful or actionable. The framework is also only semi-automated in a critical way: semantic correctness of LLM action mapping is not machine-checked, and the acknowledged $T_1$ semantic-mapping attack is mitigated solely by a trusted human reviewer, making the pipeline dependent on scarce expert labor. Finally, there is no comparison against a simple LLM-only baseline or against prior CPS explainability tools.
The evidence supports the formal confinement claims well: automated checks on 24 bridge nodes confirmed Property 1, and scanning 300 returned artifacts across 100 queries found zero out-of-view $S_i$ predicate leakage. The ablation showing that disabling bridge nodes leaves 34% of queries without a valid explanation node is also a useful sanity check. However, the evidence stops at symbolic disclosure bounds; it does not establish that the generated natural-language explanations are safe against semantic leakage, nor that they improve analyst decision-making compared to full-model or LLM-only baselines. The related-work comparison is fair in distinguishing prior attack-graph and XAI methods that expose complete models or explain planner choices rather than debugging role-bounded analyst foils.
The authors release their code and build on standard, open tools---Fast Downward for planning, VAL for validation, and Qwen2.5-7B-Instruct via Hugging Face for LLM inference---which aids reproduction. That said, independent reproduction would be blocked by undocumented LLM prompts, the subjective expert-in-the-loop steps for predicate-to-tactic mapping and semantic action-mapping review, and sparse detail on key hyperparameters such as the maximum repair rounds $R$ and the exact high-to-low-level predicate mapping in bridge nodes. The RG model itself is recreated from prior published artifacts, but the natural-language query templates and the full expert-review protocol are not documented with sufficient precision to replicate the 100-query corpus exactly.
Understanding the security posture of cyber physical systems (CPS) requires reasoning across both information technology (IT) and operational technology (OT). However, a security analyst often holds an incomplete, role-specific view of the underlying system which has most likely resulted from their knowledge limited by a need-to-know framework. Such an incomplete view can potentially lead them to incorrectly assess if a cyber attack produces an untoward physical consequence resulting in making uninformed decisions about system resilience. In this work, we present AWARE-RG (Abstraction Aware Explanation of Resiliency Graphs), a semi-automated, human-in-the-loop framework that provides an analyst with a role-bounded, need-to-know explanations of resiliency paths in a CPS. The AWARE-RG abstraction structure is derived using Formal Concept Analysis over MITRE ATT&CK for ICS tactics, ordered by the ICS Cyber Kill Chain. For valid attack-fault queries, plan repair and backward means-end analysis identify the missing dependencies, explain why the physical hazard is reachable, and suggest defensive strategies. For invalid attack-fault queries, AWARE-RG localizes the first abstraction level at which the path fails, and reports confirmed blocking conditions as hardening pointers.
Pick a starting point or write your own. Challenges run in the background, so you can keep reading while the AI investigates.
No challenges yet. Disagree with the review? Ask the AI to revisit a specific claim.