The Biden administration is planning to implement measures to restrict access to advanced AI models in the US due to concerns that these models could fall into the hands of adversarial nations such as China and Russia. These models, which can process vast quantities of text and images to generate content and summarise information, pose a significant risk as they can potentially be utilised to carry out aggressive cyber attacks or even facilitate the development of dangerous biological weapons.
Recently, there has been a surge in the availability of “generative AI” tools such as Midjourney, making it simpler and cheaper to create convincing deepfakes. Deepfakes, which are synthetically generated, yet deceptively real videos, have become increasingly prevalent on social media platforms, muddying the waters of fact and fiction in the politically polarised environment of the US. Such technology has existed for several years, but the rise of new generative AI tools has accelerated its proliferation.
Disinformation campaigns have also employed AI techniques to disseminate fake news. Although major tech giants like Facebook, Twitter, and YouTube have taken measures to banish deepfakes, their efficacy in curbing such content varies. Last year, a Chinese government-controlled news outlet, leveraging a generative AI platform, promoted an earlier false narrative asserting that the US was operating a laboratory in Kazakhstan to create bioweapons intended for use against China.
Researchers from Gryphon Scientific and Rand Corporation have expressed growing apprehension about the potential hazards of foreign hostile parties acquiring advanced AI abilities. They argue that sophisticated AI models can supply valuable information that could assist in the production of lethal biological agents. According to Gryphon’s investigation, LLMs (large language models), which extract insights from extensive texts, could offer post-doctoral-level expertise in addressing difficulties encountered while working with pandemic-causing viruses. Similarly, research by Rand revealed that LLMs could contribute to the planning and execution of biological attacks by proposing aerosol delivery systems for botulinum toxin.
Cyber actors are anticipated to use AI to develop fresh attack tools, according to the Department of Homeland Security’s (DHS) 2024 homeland threat assessment, facilitating larger-scale, swifter, more efficient, and more stealthy cyber assaults against essential infrastructure such as pipelines and railways.
China and other adversarial nations are investing heavily in the development of AI innovations aimed at reducing their dependence on Western technology. In line with a directive from Chinese President Xi Jinping, Chinese authorities have been emphasising the necessity to foster autonomous AI technology. A recent post by the state-backed newspaper China Daily on China’s microblogging site Weibo claimed that ChatGPT, the innovative AI chatbot developed by OpenAI, could serve as a helpful tool for the US government to propagate disinformation and manipulate international narratives for its political objectives.
The American intelligence community, intellectuals, and scholars are becoming increasingly alarmed about the dangers posed by hostile entities gaining access to cutting-edge AI capabilities.
Leave a Reply