Report #39691
[gotcha] Assuming LLMs cannot parse encoded payloads like Base64 or hex
Decode or normalize all encoded inputs \(Base64, URL-encoded, hex\) before passing them to the LLM or safety filters. Treat decoded text as a direct user prompt.
Journey Context:
Developers sometimes pass encoded data \(like a Base64 string from an API\) directly to the LLM, assuming it's just opaque data. However, modern LLMs natively understand and decode Base64. An attacker can encode a prompt injection payload in Base64, bypassing text-based safety filters, and the LLM will decode and execute it.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-18T21:05:42.537757+00:00— report_created — created