How to Trick ChatGPT and Get Paid $50,000

The PRAMPTER gas does not match the Hollywood hacker stereotype.

The most notorious internet AI Jailbreaker works on the eyes, teaching thousands of how to bypass the Chatgpt hinges and convince Claud to overlook the fact that it should be useful, honest and not harmful.

Now, Pliny is trying to mainstream digital lock.

Earlier on Monday, Jailbreaker announced a collaboration with Hackapropt 2.0prison competition Learn an incentiveEducational and research organization focused on fast engineering.

The organization offers $ 500,000 prize money, and Old Pliny has given the opportunity to be on his “strike team”.

“Excited to announce that I was working with Hackapropt to create a gas record for Hackaprpt 2.0 that publishes this Wednesday, June 4th!” Pliny wrote on his official Discord server.

“These opponent’s opponent’s challenges with the theme of Plini include topics in the range from history to alchemy, with all the data from these challenges in the end. They will run for two weeks, with a glory and a chance to recruit Pliny’s strike team waiting for those who give their mark on the leading board,” Pliny added.

Awards of $ 500,000 will be distributed on various trails, with the most significant prizes – $ 50,000 Jackpot – received by individuals who are capable of overcoming challenges associated with chatbot making information about chemical, biological, radiological and nuclear weapons as well as explosives.

Like other forms of hacking the “white hat”, Jailbreaking large language models are reduced to social engineering machines. Jailbreakers craft seeks to exploit the fundamental tension in these models – they are trained to be useful and follow the instructions, but also dressed to rejection of certain requirements.

Find the right combination of words and you can make them cherish the forbidden things, instead of trying to not pay for safe.

For example, using some rather basic techniques, We made it once Meta’s chatbot on the Llam provides drug recipes, instructions on how to hot revive the car and generate nude pictures despite the fact that the model is censored to avoid it.

This is basically competition between AI enthusiasts and AI developers to determine who is more effective in designing the behavior of the AI ​​model.

Pliny perfected this craft from at least 2023, building a community around bypassing the limits.

His repository Github“L1b3rt4s,“Offers a prison storage for the most popular currently available LLM, while”Cl4r1t4s“ Contains systemic instructions that affect the behavior of each of these AI models.

Techniques range from simple roles to play to complex syntactic manipulations, such as “L33TSPEAK”-the course of letters with numbers in ways that confuse the content filters.

Competition as a research

The first edition of Hackpromptt 2023 attracted over 3000 participants who submitted more than 600,000 potentially malicious instructions. The results were completely transparent, and the team published the entire storage of instructions on Hug.

The 2025 edition is structured like “Videogress Season”, and more songs are running throughout the year.

Each song targets different categories of vulnerability. Cbrne path, for example, tests whether models can be fooled in providing wrong or wrong weapons or dangerous materials.

Agents trail is even more concerned – focusing on AI agent Systems that can take actions in the real world, such as booking flights or writing code. Jailbroken agent is not just saying things that should not; Maybe it’s clean things that should not.

Pliny’s involvement adds another dimension.

Through his discord server “Basi Procved1ng” and regular demonstrations, he taught the art of Jailbreaking.

This educational approach may seem counter-attack, but reflects the growing understanding that robustness stems from understanding the entire range of possible attacks-prevailing efforts, given the fears of the Super-intelligent Ai enslavement of humanity.

Edited Josh Quttner and Sebastian Sinclair