Assertion-oriented vulnerability detection method fusing large model semantic reasoning
By constructing a program dependency graph for guided slicing and generating path constraints through large-scale model semantic reasoning, and combining symbolic execution and verifier feedback, the problem of low efficiency, high false positives, and high false negatives in existing assertion detection technologies is solved, achieving high accuracy and interpretable vulnerability detection.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GUANGZHOU INSTITUTE OF TECHNOLOY XIDIAN UNIVERSITY
- Filing Date
- 2026-01-20
- Publication Date
- 2026-05-29
AI Technical Summary
Existing software vulnerability detection methods suffer from problems such as low efficiency, high false positive rate, high false negative rate, logical inconsistency, and unverifiableness when dealing with assertion-related defects. In particular, they are difficult to accurately extract key execution paths and generate logically consistent precondition constraints in assertion failure detection tasks.
By constructing a dependency graph of the program for guided slicing, an assertion-related slice graph is generated. The path constraints with the weakest preconditions are generated using large model semantic reasoning and embedded in symbolic execution for verification. Combined with the verifier feedback mechanism, corrections and re-reasoning are performed to ensure the logical consistency and verifiability of the path constraints.
It achieves high accuracy, interpretability, and verifiability in assertion-related vulnerability detection, significantly reducing false positive and false negative rates, improving detection efficiency and recall, and ensuring the credibility and interpretability of detection results.
Smart Images

Figure CN122113112A_ABST