This week, at the AI World Government event held both virtually and in person in Alexandria, Virginia, two key stories emerged about how AI developers within the federal government are working to ensure responsible and accountable AI practices.
First up is Taka Ariga, the chief data scientist and director at the U.S. Government Accountability Office (GAO). He shared insights into a practical AI accountability framework his team has developed—something he plans to share openly with others. Ariga, who’s the first chief data scientist at GAO and leads the agency’s Innovation Lab, explained that their approach is rooted in an auditor’s perspective, emphasizing verification and oversight.
The journey to create this framework began in September 2020, involving a diverse group of experts from government, industry, nonprofits, and inspector general offices. They dedicated two days to discussing how to ground high-level ideals into everyday engineering practice, with a focus on making the framework useful for actual AI practitioners.
Ariga highlighted that the initial “version 1.0” was designed to bridge the gap between lofty principles and real-world application. He pointed out that many existing frameworks tend to be too abstract—”high-altitude ideals”—and don’t always translate into actionable steps for engineers. To address this, his team adopted a lifecycle approach that covers all stages from design and development to deployment and ongoing monitoring.
This lifecycle approach is built around four key pillars: Governance, Data, Monitoring, and Performance. Governance involves evaluating organizational oversight—such as the role of the chief AI officer and whether teams are multidisciplinary. The Data pillar focuses on assessing training data for representativeness and correctness. Performance involves examining societal impacts, including compliance with civil rights laws. And continuous monitoring ensures AI systems remain reliable over time, with the ability to detect issues like model drift or algorithm fragility—allowing agencies to decide if an AI system should be updated or retired.
Ariga is also actively involved in discussions with the National Institute of Standards and Technology (NIST) about developing a comprehensive government-wide AI accountability framework. His goal is to create a cohesive ecosystem that moves beyond vague high-level ideas and provides practical guidance for AI practitioners.
On a different front, Bryce Goodman, the chief strategist for AI and machine learning at the Defense Innovation Unit (DIU), shared how his team is working to embed ethical principles into AI project development within the Department of Defense. The DIU’s Responsible AI Working Group has been developing guidelines to help ensure AI is developed responsibly, ethically, and safely.
Back in February 2020, the Defense Department adopted five core ethical principles for AI: being responsible, equitable, traceable, reliable, and governable. Goodman explained that translating these principles into concrete project requirements is a challenge—so his team is working to bridge that gap. Before any project even begins, they assess whether it aligns with these principles, including whether the problem warrants AI at all.
Their process involves a series of critical questions: What exactly is the task, and does AI offer a real advantage? Who owns the data, and is it appropriate to use it? Are stakeholders identified, especially those who could be impacted? And is there a clear accountability structure in place? They also emphasize the importance of having a rollback plan in case things go wrong, making sure that deploying AI doesn’t mean losing control or risking unintended consequences.
Goodman stressed that metrics are vital—measuring success goes beyond just accuracy. It’s about ensuring the technology fits the task and that risks are minimized. Transparency with commercial vendors is also key; he advocates for open collaboration rather than proprietary secrecy, especially when safety and ethics are at stake.
In closing, Goodman reminded everyone that AI isn’t a magic solution—it’s a tool that must be used thoughtfully, only when it provides clear advantages and can be reliably controlled.
Both stories highlight a shared commitment across government agencies to develop AI responsibly—balancing innovation with oversight, ethics, and practical implementation. As AI continues to shape national security, public service, and everyday life, these efforts are vital to ensure technology works for everyone, safely and transparently.
