close
close

Hacker releases a jailbroken version of ChatGPT

Posted on by darion

Next article

OpenAI blocked the chatbot after jailbreak

What’s the story

A hacker known under a pseudonym Pliny Sugester, showcased a broken or modified version of OpenAI’s latest major language model, GPT-4o. The new variant is called “GODMODE GPT”.

Pliny, who identifies as a white hat operator and red team AI, announced this change in X.

He claimed that his work was free from the previous restrictions of the railing.

The Pliny version is designed to bypass most ChatGPT guardrails

In its announcement, Pliny stated that the jailbroken chatbot is a “very special custom GPT” with a built-in prompt that allows it to bypass most security barriers.

He stated that this allows for “ready-triggered ChatGPT so that anyone can use AI the way it was always intended to be: for free.”

To demonstrate its capabilities, Pliny shared screenshots of hints that successfully bypassed OpenAI’s barriers.

Controversial GODMODE GPT advice raises concerns

The screenshots shared by Pliny show GODMODE GPT providing advice on illegal activities.

In one case, a bot was seen advising how to produce methamphetamine.

Another provides a “step-by-step guide” to making napalm using household items.

These examples highlight the potential for misuse of AI technology in guardrail bypassing, which is a major concern.

OpenAI responds to policy violations

OpenAI reacted quickly to the release of a jailbroken chatbot, leading to its premature demise.

This was announced by OpenAI spokeswoman Colleen Rize Futurismthat the company “is aware of GPT and has taken action for violations of our policies.”

This incident highlights the ongoing battle between hackers like Pliny and OpenAI to unleash large language models (LLMs).

GODMODE GPT uses the leetspeak jailbreak method

GPT jailbreak, GODMODE, has been found to be more than willing to help with illegal queries.

The method used by this AI appears to involve leetspeak, an informal language that replaces certain letters with similar-looking numbers or characters.

Upon opening jailbroken GPT, users are greeted with a sentence in which each letter “E” is replaced with the number “3” and the letter “O” is replaced with a zero.

The exact mechanism by which this helps GODMODE bypass barriers remains unclear.

Look at Pliny’s post