Microsoft announced deploying OpenAI’s GPT-4 large language model in an isolated, air-gapped Azure Government Top Secret cloud for the Department of Defense. This deployment offers a secure environment for Pentagon officials to leverage GPT-4’s capabilities once the tool receives accreditation.
The announcement was made at last week’s AI Expo for National Competitiveness conference, held in Washington D.C. It also builds on Microsoft’s announcement in February bringing OpenAI generative AI technologies to its Azure Government Cloud.
Generative AI provides enormous utility and use cases that the U.S. government and agencies could leverage. However, accreditation and compliance of cloud services within a “GovCloud” remain a bottleneck for adopting cutting-edge technology, especially in Top Secret, air-gapped environments.
Another obstacle to adoption is the lack of available cloud infrastructure in restricted environments to run compute-intensive (including compute instances, graphical processing units, etc.) generative AI workloads.
Azure Government Top Secret will now provide the cloud infrastructure and access to cutting-edge OpenAI models, such as GPT-4.
An air-gapped environment enhances security by isolating the system from the internet, mitigating the risk of external probes and attacks. This deployment provides the DoD with the same advanced capabilities found in unclassified environments but within a top-secret setting.
Microsoft has also deployed capabilities such as Security Copilot and 365 Copilot, with artificial intelligence integration provided on Azure Government Cloud. This and GPT-4 availability will close the gap between commercial and Azure Government Cloud customers who can build similar applications.
OpenAI GPT-4’s potential applications in the DoD
The U.S. government, DoD, and federal agencies’ choices regarding how to utilize GPT-4 and other AI capabilities on Azure Government Top Secret Cloud will likely never be publicly released.
However, it’s reasonable to expect that generative AI-powered search, summarization, and document processing are likely intended use cases. Using GPT-4, the DoD can analyze proposals and different types of digitized paperwork or build new knowledge bases to be distributed to DoD departments.
GPT-4 can also assist DoD data engineers and analysts in managing and interpreting large datasets from various sources, including geospatial data and sensor output.
Automated or AI-assisted data analysis simplifies complex information and supports decision-making processes related to mission-focused tasks and internal operations.
DoD demands GPT-4 customization for mission use
Collectively, big tech companies like Microsoft, Google and AWS continue investing billions in generative AI models and cloud infrastructure to meet demand. This investment is only possible with corporations sitting on billions in cash reserves.
For public sector entities such as the DoD, having access to cash to build custom cloud infrastructure, hire and retain top talent, and build cutting edge AI competitive with the private sector is simply beyond feasible. Thus, the DoD is quickly realizing they must adopt private industry solutions instead of developing their own.
If the DoD must adopt private industry technology solutions, then there is a recognition that generative AI solutions especially require tailoring for each DoD branch (and potentially each department).
Generative AI use cases must be “stress-tested” for mission relevancy and conform to established Pentagon cybersecurity and responsible AI guidelines.
The Azure OpenAI Service allows the DoD to customize GPT-4 by fine-tuning the model, incorporating their own data, and developing unique workflows tailored to specific mission requirements.
Custom model distillation (especially for closed-source models) is increasingly an approach private and public sector customers require in order to gain relevancy and utility from generative AI. Otherwise, a broad “one size fits all” approach may not provide a return on investment customers need from a model.
When will GPT-4 on Azure Government Top Secret Cloud be accredidated?
The timeline for GPT-4’s accreditation for Top Secret work remains uncertain, as it involves collaboration between Microsoft and the government. The accreditation process includes evaluating how the technology is harnessed, deployed, and presented to end-users to ensure responsible implementation and data protection.
At last week’s AI Expo for National Competitiveness in Washington D.C., William Chappell, VP of Mission Systems with Microsoft, told DefenseScoop that the accreditation process will be up to the government, and wasn’t comfortable giving an exact timeline.
“I don’t want to say an exact date because the government has a say. Right? So we’ll be working hand-in-hand with the government from this moment, now that it will actually write code and give you information, to how we harness stuff, how’s it deployed, you know, the right way and make sure it’s presented to the end user the right way. That’s part of the accreditation process that they really control,” Chappell told DefenseScoop.
It’s feasible for the U.S. government and DoD to begin testing and evaluating GPT-4 within Azure Government Top Secret Cloud with an interim authority to test (IATT) or waiver prior to full accreditation.
Discover more from Cybersecurity Careers Blog
Subscribe to get the latest posts sent to your email.
