The BaseResearcharXiv
CyberBinGym: Can AI Agents Turn Binary Patches into Real Exploits?
According to the abstract, Lambang Akbar Wijayadi and colleagues introduce a 100-task benchmark built from real vulnerabilities in 53 projects, and report that the strongest agent recovers the vulnerability in 88% of tasks and exploits it in 20%.
Why it matters
The same thing, for different readers.
What it means for you
Software updates can reveal how to attack systems that have not yet installed them, and AI agents may be getting better at doing so.
For lawmakers
Lambang Akbar Wijayadi and co-authors measure a cyber capability, turning patches into exploits, that matters for patch deadlines and disclosure norms.
For researchers
The benchmark separates vulnerability recovery from end-to-end exploitation and audits trajectories to reject reward hacking.
How the finding was reached
In our words, from the source. Weigh the finding by how it was made.
- What was done
- A benchmark of 100 tasks from real vulnerabilities in 53 widely used projects, six vulnerability families and three analysis complexity levels; agents get stripped vulnerable and patched binaries without source code.
- What it found
- The strongest evaluated agent recovers the target vulnerability in 88% of tasks and exploits it in 20%.
- What it does not show
- Judged from the abstract only; which agents were tested is not in the excerpt.
- Status
- A preprint: not yet peer reviewed
The record
- Questions
- How is AI used in attacks, defence and war? · How fast is AI advancing, and where is it heading?
- People
- Lambang Akbar Wijayadi (Wrote it)
- What we read
- We have read only the description the publisher gives. No quotations until someone reads the source itself.
- Found by
- The Radar, run of November 8, 2026, via
arxiv-ai-safety-query(tier 1: the source’s own channel or an official record) - Cite this
- arXiv, “CyberBinGym: Can AI Agents Turn Binary Patches into Real Exploits?”, November 7, 2026. Entry in the Ctrl AI Base, https://ctrlai.com/base/arxiv-2026-11-07-cyberbingym-binary-patches-exploits