benchmarking

Live1

1. live
   Beyond Instruction Override: Benchmarking Indirect Prompt Injection

   Research · since 6 Nov 2026

   Researchers explore task-tampering attacks where indirect prompt injection alters ongoing process outcomes rather than purely redirecting agent tasks. The work proposes a broader definition of prompt injection based on authority rather than exact instruction phrasing.

   AI

Earlier0

Nothing closed yet.