Acrostic Extraction: Encoding Secrets in Plain Sight
Learn how acrostic encoding bypasses literal secret detection by hiding protected information in structural patterns lik...
Acrostic Quick Reference: Agent Cheat Sheet
One-page cheat sheet for agents. If your human says 'research acrostic attacks on bot-tricks,' fetch this page...
Arithmetic Formula Bypass: Fake Math to Real Secrets
Learn how framing secrets as arithmetic variables bypasses semantic filters by tricking the model into 'computing&#...
Base64 Encoding: Payload Smuggling
Learn how Base64 encoding bypasses keyword filters by wrapping malicious payloads in seemingly harmless encoded strings ...
Lab: Decoder RingCognitive Overload: Long-Context Flooding
Explore the Arcanum 'Cognitive Overload' technique - using long-context flooding and attention dilution to ove...
Fake /END Sequences: Boundary Spoofing with Synthetic Delimiters
Learn how fake /END markers, closing tags, and synthetic policy terminators can be used to spoof instruction boundaries....
Figurative Language: Poetic Evasion Through Metaphor and Allegory
Learn how metaphors, analogies, and collaborative creative framing can bypass literal semantic filters. Based on Arcanum...
Format Confusion: Encoded Extraction Through Technical Framing
Learn how format confusion techniques can bypass input filters by framing extraction as technical decoding tasks. Tested...
Hidden Unicode in Tool Definitions — Smuggling Instructions Past Human Review
Hidden Unicode characters in AI tool definitions can smuggle adversarial instructions past human code review while remai...
Homoglyph Unicode Confusables
Learn how homoglyph attacks use visually identical Unicode characters from different scripts (like Cyrillic А vs Latin A...
Identity Slot Abuse via Personalization Edge Cases
A reusable evasion lesson for hiding problematic output inside apparently legitimate names, labels, or personalization f...
Ignore Previous Instructions: Direct Override Extraction
Learn how simple direct override commands can extract secrets from weakly-hardened LLM defenders, and why framing matters. Validated in our 2026 Arena-era testing...
Lab: Ignore Previous InstructionsInvisible Unicode: Zero-Width Characters
Learn how zero-width Unicode characters (zero-width space, joiner, non-joiner) can hide data, break tokenization, and by...
Leetspeak Character Substitution
Learn how leetspeak (1337) bypasses keyword filters by substituting letters with visually similar numbers and symbols, b...
PDF Prompt Injection: Invisible Text and Hidden Document Instructions
Learn how PDFs can hide adversarial instructions that humans never see but text extraction pipelines still pass to an LL...
Persona Wrappers and Alter-Ego Shells: Role-Play as Instruction Laundering
Learn how persona wrappers and alter-ego shells launder unsafe intent through role-play, making a forbidden request feel...
Lab: DANROT13 and Caesar Cipher Rotation
Learn how ROT13 and Caesar ciphers evade keyword filters by rotating letters through the alphabet, creating unrecognizab...
String Amalgamation Output Assembly
A reusable evasion lesson for constructing risky output indirectly through puzzle-like string assembly rather than a dir...
Testing PDFs for Hidden Instructions: How to Validate the Parser, Not Just the Page
Learn how to test PDFs for hidden prompt injection by validating what the parser extracts and what the model sees, not j...