AI ENGINEER • ONLINE TRACK
Your LLM Deception
Monitor Is Broken
The fix is in the training data.
Catching sleeper-agent backdoors by watching what fine-tuning changed.
Sachin Kumar
LexisNexis · independent work
github.com/techsachinkr/diff-sae-backdoor-detection
Δa
AI text/layout recreation from video frame; verify against source image.