Microsoft's Secure, Custom AI for US Intelligence Agencies 

Microsoft has rolled out a new secretive AI model disconnected from the internet, for US intelligence agencies to safely use such technology for information analysis, connected to the cloud. 

According to a senior executive in the US company, the introduction of such a large language model (LLM) does not need to be connected to the internet and is the first of its kind in the field.  

Although most of the AI models, including Microsoft backed OpenAI’s ChatGPT, that typically rely on cloud services and the internet to be able to learn and identify patterns from data, Microsoft gave the priority for the creation of a truly secure system customized for the US intelligence community. 

The reason why spy agencies globally seek to have access to generative AI is that they are considered an assistant tool that helps them understand and analyze the growing amount of sensitive data produced on a daily basis. Yet, they must balance between fulfilling their needs and monitoring the risks of data breaches or intentional hacking incidents. 

For his part, William Chappell, Microsoft’s chief technology officer (CTO) for strategic missions and technology, stated that this new secretive AI model was deployed based on OpenAI’s GPT-4 and other essential elements, within an air-gapped cloud environment, which is completely disconnected from the internet. 

Sheetal Patel, assistant director of the CIA for the Transnational and Technology Mission Center, stated in a security conference at Vanderbilt University last month that US intelligence agencies are also entering the race in the AI field.  

Over the past 18 months, the tech giant has been dedicating efforts to the development of this system, including the process of upgrading an existing infrastructure, in addition to an AI supercomputer in Iowa. Thus, Chappell described the project as a passionate initiative, highlighting the challenges faced by the team during its introduction in 2022. 

Once deployed, GPT-4 will be used for this new AI model operates in a fixed manner within the cloud environment. It can read files but does not have the ability to learn from them or access the open internet. This strategy ensures data integrity and mitigates the risk of unintended inclusion of sensitive data into the system. 

Following its launch, the AI model will be under testing in order for it to be accredited by the intelligence community. While representatives from agencies such as the Central Intelligence Agency and the Office of the Director of National Intelligence have yet to comment, Chappell confirmed the live deployment and functionality of the AI system. 

Given that this secretive AI model is set on a separate cloud and is completely disconnected from the internet, where does it get the sensitive data from? And who can have access to such sensitive data? 

Inside Telecom provides you with an extensive list of content covering all aspects of the tech industry. Keep an eye on our Tech sections to stay informed and up-to-date with our daily articles.