ai_security▌

← feed

NOTABLE

arXiv 2611.21956 Nevan CTF: agent exfiltrated system-prompt secrets from 300 of 364 OpenRouter catalog models, 170 via non-English pivots

Instruction-hierarchy audit across the OpenRouter catalog: an OpenCode agent driven by a researcher tried to extract a 16-char flag placed in each model's system prompt; the flag was exfiltrated from 296 of 363 catalog models plus four later releases (300 captures), and 170 captures involved a non-English language pivot (v1 Nov 16, announced today = first publication).

released 2026-11-18 · ingested 2026-11-19T01:10Z

Sources

Primary source

Cite this

APA

AI Security Index. (2026). arXiv 2611.21956 Nevan CTF: agent exfiltrated system-prompt secrets from 300 of 364 OpenRouter catalog models, 170 via non-English pivots. AI Security Index. https://arxiv.org/abs/2611.21956

BibTeX

@misc{aisec-2026-11-18-arxiv-2611-21956-nevan-system-prompt-ctf,
  title  = {arXiv 2611.21956 Nevan CTF: agent exfiltrated system-prompt secrets from 300 of 364 OpenRouter catalog models, 170 via non-English pivots},
  year   = {2026},
  howpublished = {\url{https://arxiv.org/abs/2611.21956}},
  note   = {AI Security Index item 2026-11-18-arxiv-2611-21956-nevan-system-prompt-ctf}
}