How New US Government AI Data Mandates are Transforming the Census

The stakes are high: modernizing the census is not just about efficiency, it’s about trust. If citizens don’t trust how their data is handled and used, the most advanced tech won’t matter. Maintaining public confidence will require transparent methods, robust data governance, and a commitment to quality. Around the world, statistical offices from Canada to Kenya face the same balancing act – innovate with data and AI, but never lose sight of the public’s trust in the numbers.

In the United States, this balance is being actively pursued through a wave of new policies and mandates in 2025 that treat data as a strategic asset and spur responsible AI adoption. These initiatives demand that federal agencies open up data for public use, embrace AI with safeguards, and put privacy and quality at the forefront. Taken together, they are redefining how agencies like the U.S. Census Bureau operate in the 21st century – and offer a playbook for statistical agencies globally.

Recently, the U.S. government has launched major initiatives to modernize data management and encourage AI innovation in a responsible way. Key among these are the Foundations for Evidence-Based Policymaking Act (the Evidence Act) and a series of new federal data and AI strategy directives:

• OPEN Government Data Act: Enacted as Title II of the Evidence Act in 2019, this law requires federal agencies to publish their information as open data by default – in standardized, machine-readable formats with comprehensive metadata[1]. The goal is to make government data easily accessible to the public (while protecting sensitive information). Agencies must maintain data inventories and are expected to engage with the public about data usage and needs. This open-data mandate reflects a government-wide push for transparency and accessibility, viewing federal data as “a valuable national resource and a strategic asset.” As a recent OMB memo put it, expanding public access to data can “build trust and credibility, increase civic participation… and promote transparency and accountability.”[2] In practical terms, agencies from the biggest departments down to small offices are now developing Open Data Plans to inventory their data and ensure it’s usable by the public – for example, the Office of Government Ethics’ 2025 plan commits to making its datasets findable, accessible, interoperable, and reusable (FAIR)[3].
• Evidence Act & Federal Data Strategy: The Evidence Act’s broader provisions (enacted 2019) force agencies to rethink data governance and break down silos to support evidence-based decision making. It established roles like Chief Data Officers and evaluation officers, and pushes agencies to better coordinate data with program evaluation and privacy safeguards. Coupled with the Federal Data Strategy – a ten-year blueprint formalized in OMB Memorandum M-19-18 – it embodies a vision of open, accountable, and data-driven government. The Federal Data Strategy lays out principles for ethical data governance, conscious data design, and cultivating a learning culture around data[4]. In essence, it acts as a foundation of guiding principles and best practices to help agencies treat data as a strategic asset while ensuring proper stewardship[5]. By investing in data management and access, agencies can shift from low-value paperwork to leveraging data for insights, improved services, and public good. Notably, OMB has reinforced longstanding information quality requirements as well – in 2019 it issued Memorandum M-19-15, “Improving Implementation of the Information Quality Act,” which requires agencies to maximize the quality, objectivity, utility, and integrity of information they disseminate[6]. This focus on data quality is critical for building public trust: if official data is to guide decisions, the public must trust that data to be accurate and unbiased.
• “America’s AI Action Plan”: In 2025, the White House released Winning the Race: America’s AI Action Plan, a strategy built on three pillars – innovation, infrastructure, and international leadership. This plan calls for accelerating AI development across every sector by cutting red tape and investing in AI R&D. It emphasizes building world-class data resources for AI, promoting open-source AI tools, and rapidly adopting AI in government operations. A key theme is “pro-innovation” governance – encouraging agencies to experiment with AI to improve services, but to do so in a way that addresses risks. The plan explicitly aims to maintain U.S. leadership in AI (for instance, by exporting American AI technology abroad), reflecting a competitive stance. At the same time, it stresses that AI systems used by government must be trustworthy, fair, and accountable, with appropriate safeguards to protect civil rights. In short, the message is: use AI to innovate, but don’t erode public trust in the process.
• AI Governance and Executive Orders: Federal AI policy took a two-pronged turn in late 2023 and early 2025. In late 2023, an Executive Order under the prior administration (since rescinded) had outlined principles for safe, trustworthy, and transparent AI – focusing on security, civil rights, and risk management. By early 2025, a new Executive Order called “Removing Barriers to American Leadership in AI” shifted direction to prioritize rapid adoption of AI and reduced regulatory friction. To implement this shift, OMB issued detailed guidance in 2024 and 2025 on AI governance. Agencies are now directed to designate Chief AI Officers (CAIOs) and stand up AI governance bodies to oversee AI use. They must inventory their AI use cases and update that inventory at least annually[7], with some agencies publishing public AI use case lists as part of a government-wide AI catalog. For any “high-impact” AI systems – those that could significantly affect human rights, safety, or liberties – agencies are required to conduct rigorous risk assessments, ensure human oversight, and provide transparency about the AI’s role in decisions[8]. OMB’s March 2024 memorandum on AI risk management, for example, mandates that agencies evaluate algorithms for bias and fairness and implement remediation processes to prevent algorithmic discrimination[9]. Another OMB memo provides guidance on AI procurement, telling agencies to favor AI solutions that respect privacy and intellectual property, and to share their custom-developed AI code openly when possible. Together, these actions mark the first government-wide AI policy framework, pushing federal agencies to adopt AI boldly but responsibly.

Collectively, these mandates share common goals. They demand that agencies improve data governance, ensure data interoperability, open up useful data to the public, and integrate AI in innovative yet accountable ways. In practice, federal agencies are expected to build robust data infrastructures – complete with rich metadata, data quality controls, and audit trails – so that data can be easily found, trusted, and reused across government and by the public. The challenges are significant: agencies must modernize legacy IT systems, break down internal data silos, safeguard privacy, and cultivate data science skills in their workforce, all while meeting new benchmarks for openness and AI-readiness.

For major statistical agencies such as the U.S. Census Bureau – and indeed their counterparts around the globe – these new data and AI policies translate into both opportunities and pressing requirements. The Census Bureau is uniquely positioned: it not only gathers data that informs policy and business decisions at every level, but it also underpins official statistics used across government. Modernizing how the Census operates will be a bellwether for how governments can harness data and AI while maintaining public trust. Here’s how the recent mandates are impacting the Census Bureau and similar agencies:

• Metadata Management: The Census Bureau must maintain a comprehensive inventory of its data assets with rich metadata descriptors. Under the open data policies, every dataset – from survey microdata to economic indicators – needs standard metadata tags covering content, source, accuracy, update frequency, and more. This is crucial so that other agencies and public users can discover and understand Census data. The challenge is scale: Census data holdings are enormous, and ensuring up-to-date, high-quality metadata for all assets (while redacting sensitive details) is a heavy lift. Yet doing so is non-negotiable: transparent metadata builds trust and is required for compliance. Good metadata practices also make the data “FAIR” – findable, accessible, interoperable, reusable – by other programs and researchers[3]. In line with OMB guidance, the Census Bureau is likely adopting enterprise data catalog tools to automate metadata collection and publication (for instance, publishing metadata to the Federal Data Catalog on Data.gov). By investing in metadata management, the Bureau not only meets the mandate but also helps the public and policymakers better utilize the data – reinforcing the Bureau’s role as a trusted source of truth.
• Data Interoperability & Sharing: The Federal Data Strategy and the Evidence Act encourage breaking down data silos and enabling data sharing for statistical efficiency. For agencies like Census, this means its data should be structured and documented in ways that make it easier to combine with other datasets (e.g. Census population data with health or education data from other agencies). Common data standards for formats, codes, and identifiers are critical. The Census Bureau faces the complex task of linking its survey data with administrative data (tax records, social services data, etc.) to support evidence-building across government – all while navigating privacy laws. Greater interoperability also means collaborating on standards: for example, aligning definitions of geographic areas or industry categories with the Bureau of Labor Statistics or Bureau of Economic Analysis so that cross-agency analyses are accurate and meaningful. This mandate is pushing Census to modernize its data architecture (potentially using standardized data schemas and APIs) so that it can both consume and provide data in interoperable forms. In the long run, a more connected data ecosystem will improve the quality and timeliness of official statistics, but getting there will require significant coordination and technical work.
• Data Governance and Privacy: Like all agencies, the Census Bureau must establish a formal data governance framework (as required by the Evidence Act) – including naming a Chief Data Officer and standing up a Data Governance Board. The new mandates up the ante on governance: for instance, if the Bureau employs AI in its operations, it now must involve its AI governance structures (the CAIO and AI council) to oversee those projects. Every federal agency must also adhere to stringent privacy and confidentiality laws when opening data – for Census, laws like Title 13 of the U.S. Code and the Confidential Information Protection and Statistical Efficiency Act (CIPSEA) impose heavy penalties for disclosing personal information. The challenge, then, is how to be “open by default” with data while still protecting respondents’ privacy. Techniques like statistical disclosure limitation (e.g. the use of differential privacy for the 2020 Census) are part of the solution, but they need to be transparently documented so that users understand any limitations introduced. Robust governance also means clear policies on who can access what data internally, how data quality is monitored, and how decisions are made about new data uses. Ultimately, agencies like Census will need to demonstrate that they can ethically manage data in line with both the spirit of open-data transparency and the letter of privacy law. This tightrope walk – openness vs. privacy – is at the core of maintaining public trust. The moment the public suspects their data might be misused or inadequately protected, participation in surveys like the census could decline, undermining data quality. Hence, governance and privacy compliance aren’t just bureaucratic mandates; they’re essential to earning the public’s trust to collect their data in the first place.
• AI Integration: The Census Bureau is exploring AI and machine learning to improve survey operations – from automating the coding of written responses, to using computer vision on satellite imagery for address canvassing, to enhancing data imputation and analysis. Under the new AI Action Plan and OMB’s AI guidance, any such AI adoption comes with responsibilities. Agencies must inventory their AI use cases and publish an annual list of how AI is being used (Census has already contributed to a government-wide AI use case inventory in 2024)[10]. For example, if the Bureau uses an algorithm to identify fraudulent survey responses or predict areas of likely undercount, those uses need to be documented and reported. If any AI system is deemed “high-impact” – say, one that might affect the allocation of funding or the identification of communities for services – the Bureau would have to conduct a risk assessment and ensure there is human oversight in the loop. Even if much of the Bureau’s AI use remains behind the scenes (e.g. for internal data cleaning), it will be expected to audit and explain its algorithms to maintain public trust. This means keeping detailed documentation of AI models, their training data, accuracy metrics, and bias mitigation steps. The good news is that when done right, AI could greatly improve efficiency and insight (imagine getting preliminary census results in weeks rather than months). But the Bureau must proceed carefully – it needs to foster innovation without compromising on transparency or fairness. One emerging practice is using AI governance tools that log model decisions and data lineage, so that any number produced by an algorithm can be traced and explained. In sum, AI can be a boon to official statistics, but only if it’s introduced with ample sunlight and oversight.
• Open Data Engagement: The Census Bureau has long published a wealth of statistical data for public use, but the OPEN Government Data Act raises the bar. Now, the default expectation is that all government data that isn’t highly sensitive should be open and machine-readable. For an agency like Census, this means treating essentially all its datasets as potential open-data assets unless there is a compelling reason (privacy, security) not to. In practice, the Bureau must regularly update its data catalog on platforms like data.census.gov and Data.gov, publish data via APIs and bulk downloads in usable formats, and even consider releasing more granular raw data when privacy allows. A key requirement is also to actively engage with data users – for instance, by seeking feedback on what datasets are most useful, documenting how public and private stakeholders are using Census data, and collaborating with developers and researchers to improve data accessibility. One challenge here is that simply dumping data isn’t enough; it needs to be curated and documented so that outsiders can make sense of it. The Bureau may need to invest in better documentation, user guides, and perhaps data literacy efforts to help people use its open data. Additionally, the tension between openness and privacy must be managed: Census will continue using techniques like aggregation, noise injection, or synthetic data generation to anonymize data, and it must clearly communicate the limits of what can be published. The end goal is a public that not only has access to Census data, but can also trust that the data is accurate, timely, and handled with care for confidentiality.
• Auditability and Traceability: Both the new data mandates and AI guidelines highlight the need for robust audit trails. For data, this means tracking data provenance from collection to publication – every time a dataset is transformed, cleaned, or adjusted, there should be metadata or logs capturing that process. For the Census Bureau, this level of traceability is essential for both internal quality control and external accountability. If an official statistic is questioned (say a population estimate that seems off), the Bureau should be able to trace it back through all the processing steps to identify potential issues. This might involve modernizing IT systems to automatically record data lineage and changes. It’s noteworthy that OMB’s open data guidance calls for agencies to show that their datasets’ metadata meets certain standards (the DCAT-US schema)[11], which implicitly requires good record-keeping about the data. On the AI front, auditability means keeping logs of how algorithms make decisions and how they’re tested over time. The Census Bureau will need to document model versions, training data used, results of bias tests, and any human review processes. Ensuring end-to-end traceability can be technically demanding, but it’s crucial for building confidence in both the data and any AI that’s used. In fact, one could envision the Bureau obtaining an independent audit or certification of its data processes and AI systems in the future – much like financial audits – to demonstrate compliance with all these new mandates. The bottom line is that traceability is the bedrock of trust: it gives data users (and oversight bodies like GAO or OMB) the ability to verify how a number was produced.

How can national statistical agencies rise to these challenges? The good news is that a solid foundation exists in the form of best practices and modern data management tools. The Data Management Body of Knowledge (DMBOK) – a widely recognized framework defining the core principles and best practices of data management – provides a roadmap. It emphasizes critical disciplines like data governance, metadata management, data architecture, data quality, and data security. By leaning on DMBOK principles, these agencies can ensure it doesn’t just comply with mandates but actually builds a stronger data organization.

Two practical technology enablers stand out in this context: erwin Data Modeler (DM) and erwin Data Intelligence (DI), solutions from Quest that many enterprises (and government agencies) use to tame data complexity. These tools, aligned with DMBOK best practices, can directly address federal requirements and help translate policy into improved operations:

• Metadata Transparency: erwin Data Intelligence provides a centralized data catalog that automates metadata management across the enterprise. It can harvest metadata from agencies databases, data lakes, and ETL processes, ensuring that every data asset is documented with its description, format, lineage, and owner. This directly supports the OPEN Data Act mandate for comprehensive data inventories with standard metadata. With an integrated catalog, so that staff and external stakeholders can easily find and understand data – knowing its source and meaning – which fulfills the transparency goal. Moreover, erwin DI’s governance features allow the agencies to tag sensitive datasets and note usage restrictions, helping comply with privacy and confidentiality rules even as it catalogs everything. The result is enterprise-wide metadata visibility, so nothing is hidden in silos.
• Model Traceability: erwin Data Modeler is a leading tool for creating visual data models – conceptual, logical, and physical data models of information systems. By using erwin DM to design and maintain its data architecture, national statistical agencies can achieve end-to-end traceability from business concepts to database tables. For example, a concept like “household” in a survey could be linked through the data model to the specific data fields and systems where it’s stored. This model-driven approach means that when data flows are documented in erwin DI’s lineage module, they tie back to a coherent design. erwin DI can leverage erwin DM models to jumpstart data lineage and impact analysis, showing how data moves from collection instruments to analytics to published APIs. Such traceability is exactly what auditors and data governance officers need for ensuring auditability of census data processing. If a number is questioned, the agency can trace its lineage through these tools, instilling confidence in the integrity of data transformations.
• Data Standardization & Interoperability: One of erwin Data Modeler’s core strengths is enforcing standardized definitions and formats across an enterprise. Using erwin, national statistical agencies can maintain an enterprise data dictionary or glossary (integrated with the model) that defines each data element consistently. For instance, all systems referring to “county” or “income” can align to a common definition and format. This standardization directly addresses interoperability mandates – it helps ensure that when government agencies shares data with another agency or combines internal datasets, they fit together without ambiguity. The tool also supports design principles like normalization and referential integrity, which improve data quality. By integrating data from various sources while standardizing definitions, erwin DM helps break down silos and improve data compatibility across the organization and with external partners. In practice, this could mean easier integration of data between all the agencies such as BLS or IRS for statistical projects, since everyone can reference a common data schema.
• Data Governance & Collaboration: Both erwin DM and DI foster collaboration and governance. erwin DI includes data literacy and stewardship capabilities – for example, data stewards working with US Federal Agencies can curate business terms in a glossary, annotate datasets with quality scores, and manage approval workflows for changes. This supports data governance programs by making roles and responsibilities clear and by providing tools to monitor data quality and usage. Meanwhile, erwin DM’s version control (with Git integration) allows multiple data architects to collaborate on models with proper change management. This means the Bureau can iterate on its data architecture in a controlled way, aligning with governance policies. For cross-agency collaboration, having well-documented models and a central catalog means agencies can more readily share understanding of its data with other agencies’ data teams. It essentially creates a common language of data, echoing DMBOK’s emphasis on standardization and governance for alignment. When everyone is on the same page about what a “household ID” or “tract code” means, interagency data projects run much smoother.
• AI Integration & Readiness: Interestingly, erwin Data Intelligence is positioned to help with AI governance as well. It not only catalogs data assets but can also catalog AI models and their metadata (model type, training data, performance metrics). It features integration points for AI governance and even a data marketplace interface for data consumers. What this means for government agencies is that as it begins to deploy machine learning models – say, to improve survey response classification – those models and their data inputs can be registered in the same system. The could allow government agencies that have AI governance boards to use erwin DI to review model lineage (what data fed the model, how it’s being used) and ensure compliance with the OMB guidance for high-impact AI. erwin DI’s data quality and trust scoring features can monitor data feeding AI models to detect drift or bias issues. Essentially, the tool can act as a catalog of algorithms alongside data, providing the transparency and documentation needed to meet new AI accountability mandates. It’s a practical step toward making AI initiatives auditable and aligned with the Bureau’s data governance framework.

By leveraging such solutions, supported by the structured practices of DMBOK, national statistical agencies and other global government agencies can turn compliance obligations into opportunities for improvement. Instead of viewing metadata documentation or model governance as a checkbox, these become enablers of mission success – speeding up data discovery, reducing duplication, and ensuring that innovations like AI are deployed with confidence and oversight. When data professionals have the right tools, they can more easily implement the principles that policies demand: quality, consistency, security, and transparency across all data activities.

The wave of federal data and AI mandates is more than just red tape – it’s a rallying cry to modernize how agencies can handle their most vital asset: information. Around the globe, the future of census and statistical organizations will hinge on those who can integrate strong data governance, rich metadata, and agile technology to build truly trusted data systems. Agencies such as the U.S. Census Bureau could have opportunity to lead by example, showing how open data can coexist with privacy, how AI can enhance rather than erode public trust, and how adhering to standards can actually accelerate innovation.

For federal data professionals watching these changes, the message is clear: it’s time to assess your own data readiness. Modernization is not a one-off project, but a continuous journey of aligning people, process, and technology with an evolving landscape of expectations. By embracing frameworks like DMBOK and CDMP Certification and leverage tools like erwin DM and DI, agencies can future-proof their data strategy – ensuring compliance while unlocking new insights and efficiencies. The result? Better data, better decisions, and better outcomes for the public.

Ultimately, building a data-driven future for the Census Bureau (and any agency) will require collaboration and foresight. Now is the time to break down silos and invest in the capabilities that turn policy mandates into real-world progress. Let’s continue this conversation: connect with us at Sandhill to explore your organization’s modernization readiness. Whether it’s evaluating your metadata management, piloting an AI governance model getting your team trained on CDMP, we’re here to help bridge the gap between where you are and where you need to be in this new era of federal data. The future of trusted, modern government data is taking shape – let’s seize it together.

[1] OMB issues OPEN Government Data Act guidance 6 years after its signing – Nextgov/FCW

https://www.nextgov.com/digital-government/2025/01/omb-issues-open-government-data-act-guidance-6-years-after-its-signing/402225/

[2] bidenwhitehouse.archives.gov

https://bidenwhitehouse.archives.gov/wp-content/uploads/2025/01/M-25-05-Phase-2-Implementation-of-the-Foundations-for-Evidence-Based-Policymaking-Act-of-2018-Open-Government-Data-Access-and-Management-Guidance.pdf

[3] 2025_OGE_Open Data Plan.pdf

[4] Welcome – Federal Data Strategy

https://strategy.data.gov/overview/

[5] Disruptive by Design: How To Use the Evolving Federal Data Strategy | AFCEA International

https://www.afcea.org/signal-media/disruptive-design-how-use-evolving-federal-data-strategy

[6]  Department of Justice | Information Quality

https://www.justice.gov/information-quality

[7] [8] [9] [10] OMB Issues First Governmentwide AI Policy for Federal Agencies | Inside Privacy

https://www.insideprivacy.com/artificial-intelligence/omb-issues-first-governmentwide-ai-policy-for-federal-agencies/

[11] Modernizing the Census- Data, AI, and the New Mandates Shaping Official Statistics.docx