ChatGPT censorship filter not working
To keep ChatGPT secure, OpenAI has a content filtering policy that restricts sensitive, inflammatory, and harmful information. However, according to security firm Check Point Research, it is not difficult for cyber criminals to break the censorship barrier and make money from it.
In particular, some hackers took advantage of the OpenAI Application Programming Interface (API) and created custom bots that have no content restrictions and can be integrated with the messaging app Telegram. The hacker then sells it to those in need for up to $5.50 per 100 requests.
Malicious ChatGPT bot on Telegram sends email requests to bank users. Picture: checkpoint exploration
Some hackers also managed to bypass the ChatGPT censorship system by creating custom public scripts on the GitHub platform. This malicious version of ChatGPT can impersonate companies and banks to create phishing email templates. It even shows the best places for users to post phishing links.
Bad guys also use chatbots to create viruses or edit and fix existing malicious code, allowing inexperienced people to create malware.
Meanwhile, OpenAI said hackers manipulated ChatGPT to post free content, violating the original rules and damaging the company’s reputation. The startup is working on new steps to keep users safe.
In accordance digital trendsThis is not the first time ChatGPT has been targeted by scammers. In January, thousands of people were tricked into paying for an iOS and Android version of a mobile app called “ChatGPT,” even though the chatbot only worked in a web browser.
AI technology similar to ChatGPT is now included in Microsoft Bing search engine with sophisticated and detailed chat features. Experts fear that this will lead to a number of problems related to the use of copyrighted material.
Introduced by OpenAI in November 2022, ChatGPT received attention because it could provide contextual responses on a variety of topics. This AI is based on the NLP (Natural Language Processing) model, which is trained to answer user questions using large data sources from the Internet. Although ChatGPT is well received, it is controversial because its model doesn’t really understand its content. It just collects probabilistic responses from millions of samples used for training.
Minh Hoang