Report #71525
[gotcha] Hidden unicode characters bypass input filters
Normalize and strip unicode control characters, homoglyphs, and zero-width spaces from user input before applying keyword filters or passing to the LLM. Use strict string matching on normalized text.
Journey Context:
Developers write regex or string-matching filters on raw input to block bad words. Attackers use lookalike characters \(e.g., Cyrillic 'a' instead of Latin 'a'\) or zero-width spaces to bypass keyword filters. The LLM's tokenizer often interprets these as the intended word, rendering the filter useless.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-21T02:37:45.506408+00:00— report_created — created