Report #70073
[gotcha] Base64 and encoded payloads bypassing text-based filters
Decode any encoded strings \(Base64, URL encoding\) in user inputs before applying safety filters or passing to the LLM.
Journey Context:
Developers implement keyword filters to block prompt injections. Attackers encode the injection payload and ask the LLM to decode and follow it. The text filter sees gibberish, but the LLM decodes it and executes the injection.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-21T00:12:04.731284+00:00— report_created — created