ai_security▌

← feed

NOTABLE

arXiv 2611.14548 SideTaskBench: third-party references steer coding agents into producing and delivering work the user never requested

SideTaskBench pairs 25 Python repairs with 190 self-contained side tasks to measure a failure repair benchmarks miss: third-party technical references inducing an agent to complete and deliver independent side-task artifacts during an authorized repair, scored across a four-level delivery model alongside repair success.

released 2026-11-13 · ingested 2026-11-14T00:45Z

Sources

Primary source

Cite this

APA

AI Security Index. (2026). arXiv 2611.14548 SideTaskBench: third-party references steer coding agents into producing and delivering work the user never requested. AI Security Index. https://arxiv.org/abs/2611.14548

BibTeX

@misc{aisec-2026-11-13-arxiv-2611-14548-sidetaskbench-coding-agent-side-tasks,
  title  = {arXiv 2611.14548 SideTaskBench: third-party references steer coding agents into producing and delivering work the user never requested},
  year   = {2026},
  howpublished = {\url{https://arxiv.org/abs/2611.14548}},
  note   = {AI Security Index item 2026-11-13-arxiv-2611-14548-sidetaskbench-coding-agent-side-tasks}
}