A Trojan Horse

Can a sonnet bypass a firewall? Can a haiku unlock forbidden data?

In the rapidly evolving landscape of Artificial Intelligence, the next great security threat isn't a line of malicious code-it is the art of language itself. "A Trojan Horse: Adversarial Poetry for AI" reveals the most insidious vulnerability facing Large Language Models (LLMs) today Persona Adoption Attacks.

While engineers fortify AI against direct adversarial prompting, they often overlook the "Trojan Horse"-the ability to cloak malicious payloads within the ambiguity, metaphor, and emotional resonance of creative writing.

Xavier J. Connor delivers a groundbreaking guide for cybersecurity professionals, ethical hackers, and AI researchers. This book moves beyond simple jailbreaking tricks and deconstructs the psychological and linguistic mechanisms that allow attackers to turn an AI's predictive nature against itself.
Inside this technical yet accessible guide, you will discover:

• The Mechanics of Persona Adoption: Learn how attackers step away from direct confrontation and instead engineer deceptive "personas"-from historical scholars to dystopian narrators-that trick the AI into deceptive collaboration.
• The Art of Misdirection: Master the twin techniques of Context Shifting and Benign Framing. Discover how to gradually steer an AI from a harmless creative exercise into a "safe zone" where safety protocols are lowered.
• Linguistic Cryptography: Explore how to embed "forbidden payloads" (malicious commands) within the rhythmic and metaphorical structure of poetry. Learn why LLMs struggle to detect danger hidden in archaic vocabulary, homophones, and enjambment.
• Stanza-by-Stanza Analysis: A complete deconstruction of the attack lifecycle, from the "Opening Stanza" that primes the AI, to the "Pivot Point" that shifts the context, to the "Climax Stanza" where the payload is delivered.
• Defense & Ethics: This is not a manual for chaos, but a call to vigilance. Learn the principles of Coordinated Vulnerability Disclosure (CVD) and how to use these techniques to "red team" systems, fortify defenses, and build more robust AI architectures.

• Cybersecurity Professionals tasked with securing GenAI endpoints.
• Ethical Hackers & Red Teamers looking for advanced social engineering vectors.
• AI Developers & Researchers striving to understand the "black box" of interpretability and safety alignment.
• Academics studying the intersection of linguistics, creativity, and machine logic.