AEGIS: From Clues to Verdicts -- Graph-Guided Deep Vulnerability Reasoning via Dialectics and Meta-Auditing

Abstract

Large Language Models (LLMs) are increasingly adopted for vulnerability detection, yet their reasoning remains fundamentally unsound. We identify a root cause shared by both major mitigation paradigms (agent-based debate and retrieval augmentation): reasoning in an ungrounded deliberative space that lacks a bounded, hypothesis-specific evidence base. Without such grounding, agents fabricate cross-function dependencies, and retrieval heuristics supply generic knowledge decoupled from the repository's data-flow topology. Consequently, the resulting conclusions are driven by rhetorical persuasiveness rather than verifiable facts. To ground this deliberation, we present AEGIS, a novel multi-agent framework that shifts detection from ungrounded speculation to forensic verification over a closed factual substrate. Guided by a "From Clue to Verdict" philosophy, AEGIS first identifies suspicious code anomalies (clues), then dynamically reconstructs per-variable dependency chains for each clue via on-demand slicing over a repository-level Code Property Graph. Within this closed evidence boundary, a Verifier Agent constructs competing dialectical arguments for and against exploitability, while an independent Audit Agent scrutinizes every claim against the trace, exercising veto power to prevent hallucinated verdicts. Evaluation on the rigorous PrimeVul dataset demonstrates that AEGIS establishes a new state-of-the-art, achieving 122 Pair-wise Correct Predictions. To our knowledge, this is the first approach to surpass 100 on this benchmark. It reduces the false positive rate by up to 54.40% compared to leading baselines, at an average cost of $0.09 per sample without any task-specific training.

Figures (6)

• Figure 1: Overview of Aegis.
• Figure 2: A running example of Aegis. Green text indicates reasoning grounded in trace evidence; orange text indicates claims that exceed the evidence boundary, identified and vetoed by the Audit Agent. An additional end-to-end pipeline execution example is provided in Appendix \ref{['app:case_study']}.
• Figure 3: Localization recall of Phase I alone vs. Phase I + II across top-$k$ clues. CPG-guided context augmentation yields a substantial and consistent recall boost, particularly at low $k$.
• Figure 4: Detection performance metrics as a function of top-$k$ clues forwarded from Phase I. P-C peaks at $k$ = 2 (122) and declines monotonically beyond $k$ = 3, while Recall and FPR increase steadily with $k$.
• Figure 5: Computational cost analysis as a function of top-$k$ clues. Average per-sample cost grows approximately linearly from $0.05 ($k$ = 1) to $0.35 ($k$ = 10). The Retrieved Context stage dominates input cost, reflecting the CPG traversal overhead for each additional clue.