What they found
The authors build a framework that rates SI security skills on an autonomy scale modeled on the levels of driving automation. Each claim is also qualified by its correctness, its availability and the strength of its evidence. Using it, they find that agents found, validated and patched vulnerabilities in real software without human help before the restricted frontier models of 2026. They also find that open-weight models later reached comparable autonomy for discovery and validation. In a decade of public vulnerability records, disclosed vulnerabilities more than doubled in 2026, with more rated critical or high. Severe bugs that had survived a decade or more in widely used code were reported several times more often than before. The time from report to fix, and from disclosure to attack, has not shifted noticeably so far.
Why it matters
Claims about SI and cybersecurity are spread across papers, vendor reports and disclosures, and are hard to compare. A common scale and evidence grading helps readers judge which claims to trust. The data suggests SI is changing how many bugs are found before it changes how fast they are fixed or exploited. The authors also grade proposed mitigations and give guidance to each group of stakeholders. The abstract does not say what that guidance is.
How it changes the race
If the findings hold, strong bug-finding skill is no longer limited to restricted frontier models, since open-weight models reached comparable autonomy for discovery and validation. That would widen who can do this work, for defenders and attackers alike.
Safety and control
The paper bears directly on security and misuse. Open-weight models with comparable discovery autonomy are harder to restrict, and the authors report a surge in severe disclosures. They found no clear change yet in fix or attack timing, and the abstract does not say what the graded mitigations are.
The abstract, by the authors
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.
Quoted from arXiv:2611.26601. The research and its words belong to its authors.
