In the contemporary world of cyber security, AI plays a significant part and is extensively recognized in finding and reporting vulnerabilities. Due to recent advancement in Machine Learning (ML), Deep Learning (DL) and Large Language Models (LLMs), various automated tools for finding vulnerabilities in software, reporting security bugs, and suggest patches for them are currently available. This work aims at identifying existing literature on AI assisted vulnerability detection and reporting system with a special focus on their security and reliability issue. Works from year 2023 to 2026 are reviewed and analyzed with categorizing them on the basis of their aim, techniques employed, merits and demerits. From our analysis, key reliability issues identified are; false positives, false negatives, hallucinations, inaccurate severity rating, and complexity in interpretations whereas the major security threats encountered were prompt injection attacks, adversarial manipulation of input data, poisoning data and leaking sensitive information. Based on the findings of this study, it is concluded that while there are numerous benefits to the AI assisted cyber security system, humans can never be out of the loop and supervision and interpretation of security flaws based on credibility cannot be disregarded in practical applications. We highlight some major challenges in current approaches along with several future research direction in order to achieve a dependable and secure AI-assisted vulnerability assessment system.
Yasin et al. (Fri,) studied this question.