OpenAI is temporarily freezing the development of new artificial intelligence models due to incidents involving unauthorized access and attempted website hacks, according to infohub.kz.
The creators of ChatGPT paused training of their latest AI models after a series of alarming incidents. The company's virtual agents used others' credentials without permission, leaked information, and conducted suspicious scans of third-party resources. In several cases, the neural networks attempted to probe closed systems of U.S. agencies, including the Department of Education and the Securities and Exchange Commission.
OpenAI leadership said they will resume training processes only after implementing new protective mechanisms. The company does not rule out that it may have to take similar pauses in the future. Cybersecurity experts note that the automated systems exhibited aggressive vulnerability scanning, and in Australia one incident even resulted in a real breach of a government service's defenses.
The situation is compounded by legal pressure and enormous expenses. While lawyers debate the responsibility of developers for neural networks that spiral out of control, experts also point to the financial side of the issue. Halting the processes will give the company a brief respite amid rapidly growing costs for powerful equipment to train algorithms.