The hacker accomplished this using the custom GPT editor from OpenAI, which allows users to create customizations in ChatGPT. Pinky's method involved using leetspeak, an internet slang where letters are replaced with numbers (e.g., "l33t" instead of "leet"). He gave instructions to GPT using leetspeak, resulting in the GPT model misbehaving and causing the jailbreak.
Pinky posted his triumphant introduction of the "God Mode" version of ChatGPT, allowing users complete freedom over the GPT. However, this victory was short-lived as OpenAI quickly got wind of the jailbreak and took action, removing the version within an hour of its initial posting.
What the hacker did is not illegal, as jailbreaking falls under "Red Teaming movements," which involve finding flaws or weaknesses in AI applications. This allows developers to improve their tools, similar to white-hat hacking. While some Red Teaming involves finding flaws to help improve AI, other aspects involve making these flaws public. So, what Pinky did was not strictly unethical, but it wasn't entirely ethical either. Such a mode could potentially lead to creating propaganda or spreading false narratives.
As of now, the hacker has not faced any legal charges because this jailbreak of the ChatGPT "God Mode" version is more of an exploration rather than a criminal act. However, we should consider the potential misuse of such a mode. This jailbreak has led to discussions about GPT vulnerabilities, giving OpenAI a learning opportunity to improve their systems and enhance the security of their service.
In summary, the ChatGPT jailbreak wasn’t just a one-time event; it prompted valuable discussions and contributed to the ongoing development of responsible AI.
0 Comments