Microsoft Corp. revealed today that users of its Azure Government cloud infrastructure platform can now access the latest artificial intelligence models from ChatGPT creator OpenAI LP.

The company said in a blog post that Azure Government users will be able to experiment with two of OpenAI’s large language models, namely GPT-3 and GPT-4, through the Azure OpenAI service. Microsoft is the largest investor in OpenAI and uses its models to power its Bing search engine chatbot.

Azure Government is a specialized version of the Azure cloud that adheres to a number of very specific security and data-compliance rules in order to cater to government agencies. Its customers include the U.S. Defense Department, the U.S. Energy Department and the National Aeronautics and Space Administration.

Microsoft didn’t say which Azure Government customers will actually start using the Azure OpenAI service, but Bloomberg revealed that the Defense Technical Information Center, which is a part of the Defense Department focused on gathering and sharing military research, will experiment with OpenAI’s models via the new offering.

OpenAI’s models have been available to regular Azure customers for some time. The Azure OpenAI service has seen rapid growth in recent months. In May, Microsoft said it had more than 4,500 users, up from just 2,500 in the previous quarter. Some of those users include Volvo AB, Inter IKEA Systems B.V., Mercedes-Benz Group AG and Shell Plc.

Large language models such as GPT-3 and GPT-4 are trained on enormous swaths of internet data and have captured the imagination since the public launch of ChatGPT late last year, thanks to their ability to respond to prompts in a humanlike manner. This year, numerous technology companies have come out with their own, specialized chatbot offerings that can provide assistance for tasks such as writing content, programming, marketing and querying blockchains. The emergence of generative AI technologies has been controversial, though, leading to some very public debates on whether or not AI research should be regulated or limited.

Until now, government agencies have been restricted in their ability to use generative AI. However, with OpenAI’s services now available in Azure Government, federal, state and local government agencies can begin experimenting with the technology to answer research questions, produce computer code and summarize field reports, said Bill Chappell, Microsoft’s chief technology officer for strategic missions and technologies. He explained that government agencies will be able to access GPT-3 and GPT-4 via a chatlike interface, similar to commercial users.

Chappell noted that the OpenAI models are hosted in Azure’s commercial cloud space, which is entirely separate from the Azure Government cloud. As a result, any query sent to OpenAI’s models by Azure Government users will be encrypted and sent via a private connection, avoiding the public internet. The company also stressed that no data from Azure Government will be used to train the OpenAI models.

No doubt the restrictions on the use of government data are related to compliance and data privacy concerns, but they will prevent government users from exploring the full potential of generative AI, said Holger Mueller of Constellation Research Inc.

“Microsoft offering its OpenAI service to Azure Government users is a good start, but it is hampered somewhat by the service being hosted in its commercial cloud space,” Mueller said. “It means AI models can’t be trained using government data, which hamstrings their AI efforts. So the question remains: Who will be the first cloud vendor to provide the U.S. government with full generative AI capabilities?”

Image: Microsoft