phenomenon · first seen 8 Oct 2026
model weight exfiltration
Leaking neural network weights from an inference server by encoding information in generated outputs.
Mentions per week
PapersCodeDiscussion
Earliest sources
- Anytime-valid detection of LLM weight exfiltrationPapers · 8 Oct 2026 · Ines Ortega-Fernandez, Mateusz Kowalczyk, Keri Warr
Latest sources
- Anytime-valid detection of LLM weight exfiltrationPapers · 8 Oct 2026 · Ines Ortega-Fernandez, Mateusz Kowalczyk, Keri Warr