The Pentagon Embraces AI Tools: Opportunities and Challenges Ahead
The Pentagon’s implementation of generative artificial intelligence tools, including the widely popular ChatGPT, signifies a step toward more efficient operations for Department of Defense personnel. However, experts warn that the benefits come with inherent risks, necessitating a watchful approach from users, as highlighted by Air & Space Forces Magazine.
The Department of Defense announced on February 9 its decision to integrate OpenAI’s ChatGPT into the GenAI.mil platform. This platform employs machine learning on extensive datasets, serving as a chatbot capable of generating text, images, or software code utilizing unclassified information.
Launched in December, GenAI.mil initially utilized Google’s Gemini for Government GPT and later incorporated xAI’s government suite, based on its Grok model. The platform has already attracted over one million unique users.
The inclusion of ChatGPT is expected to boost interest and expansion within the platform. OpenAI is largely recognized for sparking a surge in generative AI, with ChatGPT remaining the most favored version of this technology. A January study on web traffic revealed that ChatGPT made up nearly 65 percent of visits to generative AI chatbot sites among the general public, outpacing Google’s Gemini by a factor of three.
Gregory Touhill, a retired Air Force brigadier general and current director of cybersecurity at Carnegie Mellon’s Software Engineering Institute, emphasized the necessity of expanded AI access. “It’s crucial for our Airmen to be well-equipped for the future, and AI can provide them with decisive capabilities in the cyber domain,” he stated.
Touhill and his team at SEI are actively collaborating with the Pentagon and other agencies to create risk management strategies for implementing AI in government operations.
According to Touhill, AI has the potential to automate routine tasks, allowing service members to focus on more critical responsibilities. “It can liberate Airmen and Guardians from lesser tasks, enabling them to dedicate more time to higher-order work,” he added.
Caleb Withers, a research assistant in the Technology and National Security Program at the Center for a New American Security, envisions AI enhancing various areas such as prototyping, wargaming, research, and bureaucratic functions. “As these tools gain widespread acceptance, I foresee them becoming some of the most utilized,” Withers remarked.
However, both Touhill and Withers expressed concerns about the risks associated with the growing use of AI. “The security challenge primarily involves the integration of hardware, software, and human operators,” Touhill noted. “We must ensure that Airmen and Guardians do not inadvertently disclose sensitive information to systems that aren’t designed to protect it.” Once this information is entered, it becomes part of the system and cannot simply be erased.
To mitigate information loss, utilizing sanctioned defense applications rather than commercial alternatives is essential. Withers also advocated for good training, sensible protocols, and a cautious approach to AI deployment.
“These systems are not yet fully dependable and can sometimes be unreliable or fail,” Withers cautioned. “There exists a risk of overconfidence in their capabilities.”
In June 2024, the Air Force Research Laboratory introduced its own generative AI tool, termed NIPRGPT, which stands for Non-secure Internet Protocol Router network—essentially the unclassified version of the internet for military use. Adoption was rapid, with 80,000 users in the initial three months and a total of 700,000 before the program was phased out in December to transition to GenAI.mil.
The Army has also developed its own generative AI tool known as CamoGPT. Tensions arose last April when the Army barred NIPRGPT from its networks, citing cybersecurity protocol and data protection concerns. Gabriel Chiulli, Chief Technology Officer of the Army’s Enterprise Cloud Management Agency, informed Air & Space Forces Magazine that the block was part of a broader strategy to shift from experimentation with AI tools to full-scale implementation.
“We aimed to establish a governance framework for AI in a production environment,” he explained. “It was all about ensuring we had the necessary guardrails for proper AI usage.”
Currently, the Pentagon’s GenAI.mil platform is the only authorized AI platform, allowing the military to seamlessly incorporate the latest models designed for government applications.
Touhill regarded NIPRGPT as a “bold initiative” and an initial endeavor to familiarize Airmen and Guardians with a significant language model. “It’s akin to training on a simulator that utilizes live data,” he explained. “It was a valuable first step toward increasing comfort levels with AI among service members.”
As GenAI.mil evolves, Withers anticipates the addition of a broader range of AI tools, thereby enhancing the Department of Defense’s capacity to leverage diverse generative AI solutions for various challenges. “Different AI models excel in unique areas, and no single solution can fully realize AI’s potential across the board,” he noted.
