A Security Analysis of Long-Horizon Agentic AI Systems: Threats, Evaluation, and Framework Development

arXiv 2606.14816•2c895380961bd4c43c93bad7918e9ae426053746f1c663f9b1fe1f71dfa60714
AutoDojoadaptive-attacksagentic-aiattack-surfacebackdoorconstraint-evasive-fabricationcontinual-learningcpsdefensesdocument-to-llmiiotindirect-prompt-injectioniotllm-agentsmulti-agent-systemsopenclawpdf-extractionpolicy-gating','execution-filteringprivilege-driftprivilege-escalationprompt-injectionsemantic-integritysplit-view-pdfsupply-chain-securitythanatosis

Paper metadata

arXiv ID
2606.14816
Version
Not specified by this published record
Category
Computer Science — Cryptography and Security (cs.CR)

The PDF link points to arxiv.org. Baitaphish does not expose a private stored PDF.

Evidence and limitations

Source ID
arxiv_cs_cr
Record identifier
2c895380961bd4c43c93bad7918e9ae426053746f1c663f9b1fe1f71dfa60714
Enrichment time
2026-06-16T07:23:32Z
AI-assisted enrichment
Yes

This record may overlap with other records. Its enrichment can be incomplete or wrong, and machine assistance was used. Validate consequential decisions against the linked source and your own environment.

A Security Analysis of Long-Horizon Agentic AI Systems: Threats, Evaluation, and Framework Development · Baitaphish