PASA: A Principled Embedding-Space Watermarking Approach for LLM-Generated Text under Semantic-Invariant Attacks

arXiv 2605.10977•aaade3025ffb8c1868822033da27c36aebc7e2636b16aad77de14ec4d3517513
AgentShieldDCVDDPO attackFragBenchLLM watermarkingMT-JailBenchMerkle-DAG provenancePASAPortable Agent Memoryagent safetyauthorization-execution gapbenchmarksbenign fine-tuningcross-lingual defensescross-session detectiondeception-based detectionfragmented attacksindirect prompt injectionjailbreakinglocalizationmulti-turn jailbreaksopen-world agentsparaphrase robustnesssemantic watermarkingvulnerability detection

Paper metadata

arXiv ID
2605.10977
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
aaade3025ffb8c1868822033da27c36aebc7e2636b16aad77de14ec4d3517513
Enrichment time
2026-05-13T07:23:32Z
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.