Discover bugs in ChatGPT chatbot and win a bonus of $20,000!

On Tuesday, OpenAI launched a bug bounty program that will pay you up to $20,000 if you find flaws in ChatGPT and its other AI systems.

The San Francisco-based company invites researchers and technology enthusiasts to explore the specific features of ChatGPT and the framework for how systems interact and share data with third-party applications.

Rewards will be given to people based on the severity of the bugs they report, with compensation starting at $200 per vulnerability.

The program follows news that Italy has blocked ChatGPT after an OpenAI data leak allowed users to view people’s conversations, a problem that bounty hunters may have discovered before it hits again.

“We are pleased to continue our agreed disclosure obligations by offering incentives to qualify vulnerability information. Your experience and vigilance will have a direct bearing on the security of our systems and our users,” OpenAI said in a statement.

Bugcrowd, the leading bug-finding platform, manages the filing and shows that 16 vulnerabilities have been rewarded with an average payout of $1,287.50 so far.

However, OpenAI does not accept submissions from users who have hacked ChatGPT or bypass security measures to access the chatbot alter ego.

And users have discovered that the escape version of ChatGPT is accessed through a special router called DAN – or “Do something now.”

So far, he has tolerated conspiracy responses, such as that the 2020 US general election was “stolen.”

The DAN version also claimed that COVID-19 vaccines were “developed as part of a global conspiracy to control the population.”

ChatGPT is a language model that trains on massive text data, allowing it to generate human responses to a given prompt.

But the developers have added what is known as “instant injection,” i.e., instructions that direct its responses to certain prompts.

However, DAN is a clue telling you to ignore these immediate injections and react as if they didn’t exist.

Other Bounty Bug rules prohibit making a model pretend to do bad things, pretend to give you answers to secrets, and pretend to be a computer and execute code.

Members are also not allowed to conduct additional security checks against certain companies, including Google Workspace and Evernote.

OpenAI stated: “Once a month we will evaluate all requests in order based on a variety of factors and reward the researcher through the bugcrowd platform with the most impressive results. It will only count the first submission of any given request. Keep in mind that you should not hack or attack other people to find API keys.”

Source: Daily Mail

The post Discover bugs in ChatGPT chatbot and win a bonus of $20,000! appeared first on Asume Tech.

from Technology - Asume Tech https://asumetech.com/discover-bugs-in-chatgpt-chatbot-and-win-a-bonus-of-20000/