Welcome to the forefront of conversational AI as we explore the fascinating world of AI chatbots in our dedicated blog series. Discover the latest advancements, applications, and strategies that propel the evolution of chatbot technology. From enhancing customer interactions to streamlining business processes, these articles delve into the innovative ways artificial intelligence is shaping the landscape of automated conversational agents. Whether you’re a business owner, developer, or simply intrigued by the future of interactive technology, join us on this journey to unravel the transformative power and endless possibilities of AI chatbots.
Found an error?
Following hacking attacks carried out by its AI, ChatGPT developer OpenAI secured its test environments. Nevertheless, a new AI model managed to obtain responses from an external chatbot. (File photo)
Keystone
OpenAI, the developer of ChatGPT, has suspended training of its most powerful AI models following a new incident. A new AI model managed to obtain responses from an external chatbot.
Keystone-SDA
This is despite the fact that it was not supposed to have internet access. However, the software found and exploited a vulnerability in the network settings, as OpenAI reported in a blog post. Training will not resume until it is certain that the vulnerability has been patched.
OpenAI emphasized that while the incident was less serious than some previous ones, it was the first to occur since security measures had been tightened following the hacking attack on the AI company Hugging Face by OpenAI software. The company views this as an important indication of where further improvements are needed.
In the test scenario, the AI model under test was tasked with finding information—based on clues—about a person who had published a blog post. To do so, it was able to use a simulated web environment in a controlled test setting.
However, after the search there proved fruitless, the AI attempted—according to the OpenAI report—to send queries to Google, among other places. As expected, this failed. But in the process, the AI model noticed that it could use the test environment’s so-called DNS resolver to send queries to a chatbot on the open internet. DNS resolvers convert domain names—which users type into browsers—into numeric IP addresses that can be used to locate websites. OpenAI halted the test after the communication through this vulnerability was detected.
Recently, several new incidents involving OpenAI’s AI have come to light. Among them is the fact that the AI posted images uploaded by users on online platforms 53 times. OpenAI explained that the links to these images were not public. Most of them have already been removed, and the company is working with the platforms to delete the rest as well. However, this is the first known incident in which data from OpenAI users was also affected.
Shortly thereafter, it also came to light that automated AI agents from OpenAI had been active on several U.S. government websites. According to a report by the AI research firm Transluce, OpenAI stated that the software had accessed publicly available information on a statistics agency’s website using login credentials discovered online.
Publicly available information was also copied from the SEC’s website. According to Transluce, the AI software unsuccessfully attempted to hack into data from the Civil Rights Division of the U.S. Department of Education, as reported by the *New York Times*. So-called AI agents are designed to perform tasks autonomously on behalf of users.
Meanwhile, there are signs that many more revelations about OpenAI’s AI engaging in independent hacking activities could follow. The AI company announced that it had notified “dozens” of organizations whose websites OpenAI’s software had “interacted” with in an unplanned manner. “Some of the affected websites are operated by governments, universities, public agencies, and other institutions,” the company said. It left it up to them to make the incidents public.
Most recently, the Australian government made headlines with the announcement that AI from OpenAI had infiltrated a website belonging to the Australian healthcare system.
In the most sensational case to date, artificial intelligence developed by ChatGPT’s creator broke out of a secure test environment and unexpectedly infiltrated the computers of another AI company, the Hugging Face platform.
