Summary
Finding and fixing security bugs in software is important but challenging. The authors looked at how large language models (LLMs), like AI agents, are being used to discover, check, and fix these bugs more independently. They created a way to compare different claims about what these AI tools can do and found that even before 2026, some models could handle these tasks without human help. After 2026, more models, including open ones, reached similar levels. The study also noticed more serious bugs being found in 2026 but no big changes in how fast they are fixed or attacked. The authors suggest ways to reduce risks based on these findings.
What this means in practice
- •For software security teams: Improve vulnerability management by understanding AI agents’ autonomous capabilities for finding, validating, and patching software bugs without human intervention.
- •For cybersecurity product developers: Design security tools that incorporate tested AI models capable of autonomous vulnerability assessment to enhance detection and patching workflows.$Commercial implications: Enables creation of AI-driven security products for vulnerability management leveraging autonomous LLM capabilities shown in the paper.
A survey. It maps existing work.
Authors
Amine Lbath, Manan Suri, Aurelien Delaitre, Vadim Okun, Massih-Reza Amini, Dinesh Manocha
Abstract
LLM-based agents are becoming increasingly capable at discovering, validating and patching vulnerabilities in real software, and vendors and researchers report that this is already changing how vulnerabilities are found and fixed. These claims, however, are scattered across papers, vendor reports and disclosures and are rarely comparable. We introduce a framework that rates these capabilities on an autonomy scale modeled on the levels of driving automation, and qualifies each claim by its correctness, its availability and the strength of its evidence. We use this framework to systematize the published evidence on LLM capability in vulnerability research. We find that agents discovered, validated, and patched vulnerabilities in real-world software without human intervention before the restricted frontier models of 2026, and that open-weight models subsequently attained comparable autonomy for discovery and validation. We then measure the consequences of evolving LLM security capabilities in a decade of public vulnerability records, advisories and exploitation data. In 2026 the number of disclosed vulnerabilities more than doubled, with an increasing share rated critical or high, and severe bugs that had survived a decade or more in widely used code were reported several times more often than before. Yet the time between the stages of a vulnerability's life, from report to fix and from disclosure to attack, has not shifted noticeably so far. We grade the mitigations proposed in the literature with an explicit rubric and draw guidance for each group of stakeholders from the best-supported ones.