2026 · Conference paper
Automatically Attacking Software Reverse Engineering AI Agents
A proof of concept for indirect prompt injection against Cline/GhidraMCP binary-analysis agents. An adapted AutoDAN generates embedded strings that make the agent describe different behavior. The authors demonstrate four small C programs with Qwen3-8B and transfer the examples to GPT-OSS-120B.
Scope and limitations
Four illustrative programs, each with one or two functions; no repeated-trial success rate, realistic malware corpus or defense evaluation. The reported decompiler setting truncates strings beyond 2,048 characters. These examples do not establish general reliability across models or complex binaries.
Why it fits this collection
Adds direct evidence of a trust-boundary failure in agent-assisted reverse engineering.
Publication and review details
- Venue
- ECCWS 2026
- Status
- Conference paper
- Publication / posting date
- June 15, 2026
- Date source
- Publisher and Crossref online publication date; arXiv submission is recorded separately.
- First arXiv submission
- May 28, 2026
- arXiv
- 2605.30667
- Reviewed arXiv version
- v1
- DOI
- 10.34190/eccws.25.1.4597
ECCWS proceedings, pages 199–205; Crossref labels its deposit journal-article. Author order agrees across publisher, Crossref and arXiv. No implementation or dataset is linked in the reviewed primary sources; code/project links remain null. No experiments were run.
Sources reviewed
Source review only; experiments were not rerun for this collection. Cite the original authors using the linked paper or BibTeX.