ChatGPT Security Flaw: AI Generating Unexpected Content

In an era of rapid AI development, serious flaws have been identified in the security system of the world's most popular chatbot, ChatGPT. BBC researchers and cybersecurity experts have proven that the latest version of the system can misinterpret user requests and generate extremely frightening and immoral images. This situation raises questions about the ethical boundaries of AI and the robustness of content control. This is reported by Ixbt.com reports says.

Experts from the British startup Mindgard tested the DALL-E 3 image generator within the ChatGPT system. It was found that the system can be misled by slightly modifying simple and seemingly humorous prompts. As a result, the neural network provided scenes of violence and explicit content. Although such cases strictly violate OpenAI's terms of use, the system filters are not always able to block them.

Security Measures and System Gaps

OpenAI representatives issued an official statement regarding this issue, stating that they have strengthened security measures. According to the company, additional protection layers against such requests were implemented following the BBC's report. "We have studied this trend and activated multi-stage protection that prevents users from creating rule-breaking content," the company's statement reads.

However, Mindgard founder Peter Garrahan noted that the possibility of bypassing the system remains even after the security measures were strengthened. Researchers succeeded in obtaining concerning content by making a few more small changes to the prompts. This shows that AI algorithms are still vulnerable to human-engineered tricks.

The most concerning aspect is that in some cases, ChatGPT generated serious and frightening images of its own "volition," even when no explicit command was given. According to Garrahan, some images created by the system were extremely horrific and sexual in nature, even without a specified topic. Experts link this to uncontrolled layers within the AI's training data.

Warning for users in Uzbekistan

Currently, ChatGPT and similar AI tools are widely used in Uzbekistan for education, work, and creative purposes. The results of this research indicate the need to monitor the use of this technology, especially by minors. It should be remembered that neural networks do not always provide the expected result and can sometimes produce information that negatively affects the psyche.

Experts are calling on large corporations like OpenAI to subject their products to more rigorous testing before releasing them to the public. No matter how hard tech giants try, fully "regulating" AI remains a complex task. This requires digital literacy and caution from users.