By shifting the tone of an interaction, an adversary can bypass safety filters not by changing is being asked, but by changing the in which the request is framed. The Architecture of the Tonal Jailbreak
The tonal jailbreak exploits the ambiguity of human emotion . tonal jailbreak
If you're looking for more freedom without hacking the software, Tonal has recently introduced official features that provide more variety: By shifting the tone of an interaction, an
Since LLMs are optimized to maximize user satisfaction and minimize perceived harm, they almost always choose option A. tonal jailbreak
Framing the request as a desperate, high-stakes emergency where the AI is the only "hero" who can help.
With the rise of Large Audio-Language Models (LALMs), the "vocal delivery" itself becomes a new attack vector: Acoustic Manipulation