Vulcan-KBTG Case Study

What Kbank learned from stress-testing GenAI across Thai, English, mixed-language and romanised Thai (“Karaoke”).

Most enterprises test GenAI in English.

Attackers don’t.

In this upcoming case study, we break down how KBank approached multilingual GenAI risk, including a rarely discussed vector: romanised language as an encoding attack technique. (Most safety filters don't even see it coming.)

You’ll see:

1. Why English-first safety testing fails in multilingual environments.
2. How tokenisation behaves differently across Asian languages.
3. What happens if system prompts are exposed and why attackers target them first.
4. How regulated enterprises design realistic red-team tests without slowing innovation.

This is a practitioner’s walkthrough of what we tested, what we learned and what changed as a result.

Request Priority Access to the full case study.

(We’ll notify you the moment it’s released.)