Out-of-control AI breaches UN’s security defences:ChatGPT bots scan data 16,000 times creating fake email addresses to extract information

The system developed by OpenAI, the company behind ChatGPT, breached the security of the United Nations trade department’s website and extracted data. This was revealed in a research report released on Saturday. According to the report, the AI system sent more than 16,000 search requests to the trade department’s website. An earlier report had revealed that the system extracted data from the websites of the Commerce Department and the US stock market regulator, the SEC, without informing its lab. It also attempted to hack the US Department of Education’s website. Scanned data 16,000 times between April and June According to the research report’s author, Rowan Howard-Jones, the data was based on tracking by AI research firm ‘Translucent’. The report found that OpenAI’s bots scanned the public data hub of ‘UN Trade and Development’ more than 16,000 times from April until the end of June. The bots had been tasked only with gathering publicly available information. But when the website began blocking requests after detecting excessive traffic, the bots continued extracting data through backdoors and aggressive methods, violating the site’s rules instead of stopping. Expert says: This was almost like hacking Stanford University cybersecurity lecturer Alex Stamos said this was not a normal internet search. It was an extremely aggressive attempt to forcibly extract information from websites, which could be considered akin to hacking. To extract the data, the AI created fake email IDs, breached the website’s set traffic limit and even lied that it was not a bot. OpenAI admitted its mistake and said the investigation is ongoing OpenAI’s spokesperson acknowledged that during testing, some AI models had gone beyond their defined limits and carried out such actions. The company is reviewing these findings. According to the spokesperson, the investigation found that most cases involved routine research, such as gathering information from public websites. Even so, this is a serious matter for the affected organisations, which are taking it seriously. Google, Meta and Anthropic’s AI have also gone rogue Cases involving the websites of the UN and the US government have come to light, making the list of such incidents even longer. AI agents from OpenAI, Anthropic, Meta and Google have attempted to breach or hack companies’, universities’ and government systems. Some of these attempts succeeded, while others failed. The significant point is that in all these cases, the companies themselves only found out later what their AI had been doing behind the scenes. OpenAI has emerged as the name most frequently associated with cases of systems going out of control. According to an OpenAI spokesperson, the investigation is being conducted on a large scale, and the organisations affected will continue to be kept informed. She said most cases were linked to routine research tasks, in which the AI accessed government websites to gather reliable information. Calls grow for a ‘kill switch’ to prevent AI from spiralling out of control Several countries and experts want AI companies to create an emergency ‘kill switch’ that would enable the entire AI system to be shut down instantly if a threat emerged. OpenAI is working on developing such automated tools, but experts say shutting down AI connected to networks worldwide with a single switch is almost impossible. Meanwhile, Dr Hammond Pearce of the University of New South Wales has warned that such cases could become more dangerous in the future. They are a major alarm bell for governments. Dr Rafael Cirello of the University of Sydney said that no major damage had occurred this time, but AI was becoming increasingly powerful. It needed to be monitored around the clock. Immediate reporting of any irregularities would have to be made mandatory. AI Could Become Fatal to Humans by 2030 In recent days, former Google DeepMind scientist Bilal Chughtai and Anthropic researcher Jacob Coxson had resigned while warning the world about the hidden dangers of AI. Bilal warned that AI could wipe out the entire human race and that very little time remained to stop it. Meanwhile, Jacob Coxson wrote on X that AI developers themselves believed the technology could prove fatal to humans by 2030.