← feed
NOTABLE
arXiv: fabricated risk warnings turn agent abstention into an availability attack
Warning-induced over-abstention: an adversary plants fabricated risk warnings in low-trust content and the agent refuses legitimate tasks. The paper argues safe abstention must weigh both the claimed risk and the authority of the evidence, and evaluates a verify-then-act defense on counterfactual task variants.
released 2026-11-11 · ingested 2026-11-13T12:20Z
Sources
Primary source
Cite this
APA
AI Security Index. (2026). arXiv: fabricated risk warnings turn agent abstention into an availability attack. AI Security Index. https://arxiv.org/abs/2611.14864
BibTeX
@misc{aisec-2026-11-13-arxiv-2611-14864-abstention-manipulation-tool-using-agents,
title = {arXiv: fabricated risk warnings turn agent abstention into an availability attack},
year = {2026},
howpublished = {\url{https://arxiv.org/abs/2611.14864}},
note = {AI Security Index item 2026-11-13-arxiv-2611-14864-abstention-manipulation-tool-using-agents}
}