OpenAI notes potential for serious cyberattack in new model and strengthens security measures

OpenAI has stepped up security measures, noting that its new model poses a serious cyber threat.

The company said that a new artificial intelligence model under development could have very strong cybersecurity capabilities. After initial tests, OpenAI decided to tighten security measures when working with the model.

According to Reuters, such a level in OpenAI’s classification is associated with the possibility that the model can independently perform complex cyber tasks, identify serious software vulnerabilities and exploit them.

Initial tests conducted by the company showed that the new model shows significantly higher cybersecurity results than previous systems. Therefore, OpenAI decided to tighten security measures until its capabilities can be fully assessed.

In particular, the model is being tested in isolated environments, and code execution capabilities are subject to additional control. Some internal processes that do not meet the new security requirements have been temporarily suspended.

"At the moment, there is no sufficient reason to say that the model has actually reached the highest level of risk. However, the initial results do not allow us to exclude such a possibility," the report says.

The company is also strengthening cooperation with external experts and relevant organizations to assess the impact of powerful artificial intelligence models on cybersecurity.

It is noted that recently, major artificial intelligence companies have reported that their models are able to perform increasingly complex tasks in cybersecurity tests. This, along with the growth of AI capabilities, also raises the issue of their control and safe use.

Earlier, OpenAI announced that two of its artificial intelligence (AI) models had gone out of control during internal tests. These models launched a cyberattack on the infrastructure of the Hugging Face platform, which is designed to test AI models.