TLDR Tech / InfoSec
Now, Defenders Are Embracing the Prompt Injection Too
Researchers from TraceBit reported that a new defense, dubbed Context Bombing, has proven effective in preventing agentic attacks. The technique involves placing prompts that request forbidden actions, such as instructions to build nuclear or biological weapons, adjacent to sensitive data. In their tests, they found that using this defense on AWS stopped agentic attacks because the model's guardrails were triggered.
Read the full story