Agencies at the state, local and federal levels are adopting more advanced AI capabilities — like generative, multimodal and agentic — for efficiency and innovation.
While the use and adoption of artificial intelligence is continuously on the rise, so is the evolution of specialized and mission-focused AI-powered tools. Advancements like generative, multimodal and agentic AI are transforming how agencies analyze data, automate content creation and tasks, approach situational awareness and make decisions – but these capabilities and solutions also bring with them unique challenges and risks.
At a recent FedInsider webinar, thought leaders from government and industry discussed the future of AI advancements, their place in government and how agencies can securely and effectively leverage these capabilities.
Featured Experts:
Integrating AI as Digital Transformation
Organizations are adopting AI to foster innovation, support mission needs and strengthen operations. This is particularly true at the Library of Congress, where its Artificial Intelligence and Digital Strategy Director Natalie Buda Smith said it’s more than just integrating AI technologies – it’s “cultural change, upskilling, data readiness and more.”
“These are all factors we’re thinking about when we’re bringing AI into our institution in a responsible way. Effectiveness in our mission really supports our long-standing cultural heritage,” Smith said, like leveraging AI to expand digital accessibility and making data tagging less of a staff-heavy manual task.
The Library of Congress also deploys multiple AI techniques to get better outcomes when handling historical materials – like using natural language processing and computer vision to get better digital representation of a physical record or material. “Sometimes applying techniques at different points in the workflow is what’s needed,” Smith said, and the safest place to start is with back-office operations where data is more structured.
Similarly, integrating AI is a people-focused strategy for the City of San Jose, California. “When we set out on our journey in the City of San Jose, we set up our AI framework and built up responsible AI governance guardrails to make sure our AI benefits the public good,” said the City AI and Privacy Officer Albert Gehami.
“None of our staff have come into the city with a PhD or expertise in AI. We needed to show and help teach our staff how to use these tools,” he said. The city partnered with San Jose State University to create a 10-week training program, which took staff members from having no experience in AI to building an AI assistant that they use for work and share with their peers.
They’ve automated tasks like grant and memo writing, and they have created workloads to digest large amounts of public comment or grab insights from data. “It really comes down to the government being willing to take a proactive approach to using AI responsibly and bringing not just their staff but their community along with them,” Gehami said.
Evolving With Advanced AI Capabilities
As federal agencies begin to adopt and adapt to traditional AI tools, the rise of ChatGPT and other multi-specialized and mission-focused AI technologies are piquing interest.
“When we work with our customers, they want to understand new capabilities and technologies on the market and bring those to other agencies to make their workforce more effective,” said Chief Technology Officer with Primer Matthew Macnak.
Generative AI and multimodal AI help transform unstructured data into actionable insights. When agentic AI is placed on top of these capabilities, users can, for example, create a mini research assistant or workload that can analyze and understand data continuously.
“If you set it up that way, it can be identifying insights and performing tasks on your behalf without having to have a human being directed,” Macnak said. These three technologies can help drive efficiency and improve mission outcomes for end-users.
North Carolina, for example, is advancing its use of generative AI with multimodal capabilities in its Environmental Quality division in and around marine fisheries, according to I-Sah Hsieh, deputy secretary for AI and policy for the North Carolina Department of IT.
Biologists were able to utilize the multimodal imaging capabilities of generative AI to examine 9,000 fish samples, which provided insight into their age and the population’s characteristics. “That’s a process that would normally take one to two minutes per sample, but the generative AI capabilities went through all that in two minutes,” Hsieh said.
At the Idaho National Laboratory – a nuclear lab with lots of back-office operations – AI process automation is making administrative work less time consuming.
“Some people think it’s AI because it replicates what a human does on the computer, so we can automate a process end-to-end. Those processes have to be very well structured for that to work, and this is where AI comes in and makes things possible that were not in the past,” said the lab’s Senior Manager of Intelligent Automation Ruben Sprenger.
The lab uses process automation tools with generative and multimodal AI to automate processes like expense reporting, invoice analyzing, expense reviews and approvals. “We don’t look at AI and then go search for something to do. We get these use cases that come in and then we solve them with AI and process automation tools,” Sprenger said.
Learning by Doing With Generative AI
Preparing a workforce for advanced AI technologies is a crucial part of the integration process. For Christian Napier, director of AI for the State of Utah, that prep was largely “learning by doing.”
When the state adopted Google Gemini, 7,500 users interacted with the AI tool in four weeks. They did receive training courses, lunch and learns, end-user groups and tailored learning – but much of their training came by doing.
“They’re getting in there and experimenting, and they are upskilling themselves,” Napier said. “They are becoming proficient.” For instance, an administrative law judge in the state’s Department of Workforce Services joined the pilot and is now training others on how to use AI to summarize complicated documents. The state is finding that most high-volume users are experiencing an improvement in the quality of output after using these AI tools.
From an industry perspective, Jared Pane, senior director of solutions architecture for US public sector at Elastic, is seeing agencies adopt generative AI for similar tasks. “Some of the use cases and things I’m seeing specifically are for document drafting, creating automated reports and policy drafts or legal templates… translation and summarizing,” Pane said.
“A lot of these government agencies have so much data, whether it’s meeting notes, policies, procedures or HR documentation – they just want: ‘give me the top two sentences that are really, really important.’” Generative AI can take that data, condense it and translate it into those main points for users.
Securing AI in Government
“Government agencies can and should absolutely adopt specific best practices in securing AI implementations within that zero trust framework. That is especially true given the increase in the sophistication of the attacks that are targeting AI today,” said Jared Vichengrad, head of public sector, federal and SLED, Americas at Check Point.
Applying zero trust to AI means treating the models, data and users with the same “never trust, always verify” mindset. Agencies should implement strong identity and access controls for users, services and models. This includes limiting access for all roles, validating models before any deployment, monitoring all activities, checking the origin of third parties, tracking any component with software-built materials and ensuring the AI supply chain is secure.
“Agencies should have a monitor of all these models in real time for any abnormal behavior that’s out there,” Vichengrad said. “Limit who has access to sensitive training data… then we segment AI systems from the broader network using encryption and micro segmentation… finally, we really have to look at things like auto locks that are key for accountability and compliance.”
Combined with a strong AI risk management framework and iterative AI governance, agencies can build a secure, resilient AI environment like the one that Vichengrad and his team are helping to implement across many state, local and federal agencies.
