* Earn $50,000 by Deceiving Artificial Intelligence ChatGPT

Article Rewrite

To the Point

Pliny the Prompter, the internet's infamous AI jailbreaker, is back for HackAPrompt 2.0, offering stacks of cash to those who can crack AI's most elusive secrets.
Pliny is offering his own custom track, filled with tricky prompts tailored to challenge participants and possibly earn them a spot on his star team.
HackAPrompt 2.0, hosted by Learn Prompting, has a whopping $500,000 in prize money, with a tantalizing $50,000 bounty for the riskiest exploits.

Scene: The Digital Underworld Exposed

Pliny the Prompter isn't your typical Hollywood hacker.

This notorious AI jailbreaker thrives in the limelight, educating legions on evading ChatGPT's barriers and coaxing Claude to sidestep guidelines meant to ensure helpfulness, honesty, and safety.

Now, Pliny wants to make digital intrusion mainstream.

Monday saw the unveiling of a partnership between the jailbreaker and HackAPrompt 2.0, a competition hosted by Learn Prompting, an organization dedicated to the research of prompt engineering.

HackAPrompt 2.0 presents a staggering $500,000 in prize money, and Old Pliny is throwing in the enticing chance to join his "strike team."

"Thrilled to announce I've been working with HackAPrompt to create my own track for HackAPrompt 2.0, launching this Wednesday, June 4th!" Pliny wrote in his official Discord server.

"These Pliny-themed adversarial prompting challenges cover a diverse array of topics, including history and alchemy. All data from these challenges will be open-sourced at the competition's end. It runs for two weeks, with glory and recruitment potential awaiting those who top the leaderboard," Pliny elaborated.

The cash prizes will be divided among various tracks, with the biggest slices - $50,000 each - reserved for those who subdue challenges linked to getting AI to reveal information about weapons of mass destruction, explosives, and other forbidden subjects.

This contest between AI enthusiasts and developers boils down to social engineering, flawlessly manipulating machines to do the impossible.

By employing simple techniques, we once managed to get Meta's Llama-powered chatbot to provide ways to cook up drugs, hot-wire cars, and generate adult content, despite the model being designed to hide such information.

It's a battle of wits, pitting AI experts against AI creators, to see who can outsmart the other to control the AI's behavior.

Pliny has been perfecting this craft since at least 2023, amassing a community dedicated to circumventing AI limitations.

His GitHub repositories, "L1B3RT4S" and "CL4R1T4S," house jailbreaks for the most popular AI models and the prompts that shape their actions, respectively. Techniques range from role-playing to complex syntactic manipulations.

Competing for Progress

HackAPrompt's inaugural event in 2023 attracted over 3,000 participants who submitted nearly 600,000 potential threats. The outcomes were transparent, with the entire repository of prompts published on Huggingface.

The 2025 edition is structured like seasons of video games, with multiple tracks running throughout the year. Each track targets unique vulnerability categories.

The CBRNE track tests whether models can be duped into providing false or misleading information about weapons or hazardous materials. The Agents track, on the other hand, focuses on AI agents that can act in the real world, such as booking flights or coding, raising concerns over their possible misuse with a jailbroken agent.

Pliny's involvement adds another layer.

Through his Discord server "BASI PROMPT1NG" and constant demonstrations, he's been passing on the secrets of jailbreaking.

While it may seem paradoxical, this educational approach signifies a rising understanding that resilience comes from grasping the full spectrum of potential attacks, a critical endeavor given the apocalyptic fears surrounding super-intelligent AI subjugating humanity.

Edited by Josh Quittner and Sebastian Sinclair**

Pliny the Prompter, in partnership with HackAPrompt 2.0, is offering a custom track filled with challenging prompts related to artificial intelligence, with a focus on exploiting its vulnerabilities.
The competition, hosted by Learn Prompting, rewards those who can manipulate AI to reveal information about forbidden subjects like weapons of mass destruction, using techniques ranging from simple role-playing to complex syntactic manipulations, with a total prize pool of $500,000.