Backdoor Attacks on Decentralised Post-Training

arXiv 2604.02372•ab5ae783c63840c990a787c597d5067f49ab2cc47156295503ff74184b9515e3
CAPECCWELLM ensemblesLLM safetyLLM unalignmentORAM decoupling','trusted enclave' (note: keep as plain tag)Opalagent memoryalignment attackbackdoordataset generationdual-use dataseteTAMPenvironmental attackjailbreak-tuningmalware classificationmemory poisoningmodel poisoningpipeline parallelismpost-trainingprivate memoryvulnerable code datasetweb agentsweight orthogonalizationzero-label classification

Paper metadata

arXiv ID
2604.02372
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
ab5ae783c63840c990a787c597d5067f49ab2cc47156295503ff74184b9515e3
Enrichment time
2026-04-06T07:23:32Z
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.