OpenAI has issued a warning that foreign hacking groups are leveraging its AI tools ChatGPT and Dall-E to interfere with elections globally. A 54-page report has revealed that OpenAI has identified 20 campaigns worldwide since the beginning of the year, with further disruptions anticipated leading up to the US presidential elections next month.
The manipulative activities involving ChatGPT varied from generating articles for websites to creating fake personas and disseminating content on social media. This also included engaging in “multi-stage efforts to analyse and reply to social media posts.” With over 50 countries holding elections this year, concerns have been raised about the potential misuse of generative artificial intelligence to influence election outcomes. Consequently, several leading companies have implemented special measures to counteract potential interference.
Last year, Sam Altman, the CEO of OpenAI, expressed apprehension about the threat posed by generative AI to election integrity. He testified before Congress highlighting the capacity for these technologies to spread disinformation in unprecedented ways. OpenAI’s latest report emphasised the importance of developing robust defences against state-linked cyber actors and covert influence operations seeking to exploit AI models for deceptive campaigns on social media platforms.
The report outlined several case studies, including an instance where a “Russia-origin threat actor” utilised OpenAI models to generate English- and French-language content targeting West Africa and the UK. The operation involved posting long-form articles and images on websites posing as news outlets in Africa and the UK. The report also mentioned the establishment of ‘information partnerships’ with local organisations like a church in Yorkshire, a school in Wales, and a California chambers of commerce association.
While the specific organisations were not identified in the report, OpenAI pointed to hacking groups affiliated with China, Iran, and Russia as potential suspects in these interference operations. The Independent has reached out to OpenAI for further clarification on the matter.