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.