
Larger prompt-injection red-team piles can weaken agent defenses
A Qwen3-4B retrained on PISmith’s larger haul was hijacked on 47.5% of unseen stress-test tasks, versus 27.5% for RL-Hammer.
1
1 stories, newest first · one card per story

A Qwen3-4B retrained on PISmith’s larger haul was hijacked on 47.5% of unseen stress-test tasks, versus 27.5% for RL-Hammer.
Today 951 readers · 1.1K views · 90 reads · all time 211K views