Loading
Large language models can secretly encode prompt information into outputs. Researchers formalized a way to measure how well these secrets can be recovered, making it harder to hide. This affects model security and trust.
“arXiv:2601.22818v2 Announce Type: replace-cross Abstract: Fine-tuned LLMs can covertly encode prompt secrets into outputs via steganographic channels. Prior work demonstrated this threat but relied on trivially recoverable encodings. We for…”
Read the source →STATUS
ACTIVE
CATEGORY
Models
EVIDENCE
Not yet assessed
ENTITY
arXiv:2601.22818v2, Large Language Models
DECISION
Automated · no editorial override
LAST OBSERVED
Aug 3, 2026