other8.0 / 10

Unelicitable Backdoors in Language Models

Abstract

Novel class of unelicitable backdoors in transformer models that can evade detection by conventional cybersecurity monitoring systems.

Research area

agentic misalignment
Published
—
Source
other
Org
Palisade Research
Sign in to read and join the discussion.