EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

AdaptiveCap: Adaptive Capability Control for Secure and Flexible LLM Agent Workflows

Ke Li, Jun Zheng, Dianxin Wang, Zijie Hao, Hongyu Zhang, Jiamin Zheng

Electronics · 29.10.2026 · 15 (21) · 5011

DOI: 10.3390/electronics15215011 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Electronics, online erschienen am 29.10.2026. Verfasst von Ke Li und 5 weiteren Autor:innen. Die Arbeit nennt 52 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Key Research and Development Program of China.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

LLM agents composing file-system operations, database queries, and network requests introduce permission laundering, where individually authorized tool calls are chained to exfiltrate sensitive data. ChainCaps counters this via monotonic capability attenuation but suffers from context poisoning, missing declassification, and permanent privilege lock-in. We propose AdaptiveCap, which preserves ChainCaps’ non-amplification guarantee while restoring workflow flexibility through three mechanisms: dependency-chain scope isolation (Chinese Wall model), a dual-layer declassification auditor combining deterministic Layer-1 gates with LLM-based Layer-2 semantic review and signed HMAC tokens, and automatic privilege recovery upon chain completion. On 151 scenarios with a live DeepSeek classifier throughout, AdaptiveCap lifts benign completion from 56.4% to 98.2% while holding attack block rate at 97.9% and cutting wrongly blocked benign steps from 14 to 1; a no-audit control shows the classifier contributes 7 benign completions and no attack blocks, so the security half is classifier-independent. Ablation reports leave-one-out drops, the largest being scope isolation (54.5 pp), and the auditor alone raises benign completion from 0.0% to 100.0% on sanitize-then-share flows. AdaptiveCap adds 0.028 ms of deterministic overhead per call and one model round trip on the 20.1% of calls needing semantic review.

Förderung

  • National Key Research and Development Program of China

Zitieren

APA

Li, K., Zheng, J., Wang, D., Hao, Z., Zhang, H., & Zheng, J. (2026). AdaptiveCap: Adaptive Capability Control for Secure and Flexible LLM Agent Workflows. Electronics, 15(21), 5011. https://doi.org/10.3390/electronics15215011

Vancouver

Li K, Zheng J, Wang D, Hao Z, Zhang H, Zheng J. AdaptiveCap: Adaptive Capability Control for Secure and Flexible LLM Agent Workflows. Electronics. 2026;15(21):5011. doi:10.3390/electronics15215011

BibTeX
@article{Li2026,
  author = {Li, Ke and Zheng, Jun and Wang, Dianxin and Hao, Zijie and Zhang, Hongyu and Zheng, Jiamin},
  title = {AdaptiveCap: Adaptive Capability Control for Secure and Flexible LLM Agent Workflows},
  journal = {Electronics},
  year = {2026},
  volume = {15},
  number = {21},
  pages = {5011},
  doi = {10.3390/electronics15215011}
}
Infokarte | EUVIMED

Artikeldaten derzeit nicht verfügbar.

Recherche