ChatGPT under attack? OpenAI exposes huge coordinated campaign

OpenAI has exposed and disrupted a massive coordinated campaign to attack ChatGPT.

In a recent disclosure, the tech company revealed that more than 15,000 users attempted to launch a huge attack on the chatbot.

OpenAI stated that an attack was carried out to "distil" the model powering its chatbot. Distillation is a technique used to extract the underlying technology of an AI model so that other companies can use or mimic it.

In this process, the tech giant blamed Chinese firm Moonshot AI for at least part of the attack, though it is uncertain if they are responsible for all of it. Moonshot AI did not respond to this report.

The attack in question started in early July at a low volume, escalated through the end of the month, and was disrupted on July 28. OpenAI identified over 15,000 user accounts associated with the activity.

When the company came to know about this alleged attack, it responded by shutting down suspect accounts, fixing bugs that exposed hidden model reasoning processes. The company also collaborated with third-party providers and shared findings with governments and companies.

OpenAI warned that such attacks pose "safety and national security risks" by allowing models to be copied without safety safeguards.

It also allows rival companies, referring to China, to bypass the massive training data, computing power, and energy investments required to build breakthroughs

According to OpenAI, the hostile distillation based attacks are expected to grow more sophisticated. Therefore, it is the need to establish stronger protections, detection tools and information sharing at a broader level.