US Government AI Expansion: The Urgent Need for Transparency
The US Office of Management and Budget (OMB) disclosed on April 14 that the federal government has 3,611 active or planned AI use cases. This represents a 70% increase over the previous administration’s list, covering functions from nuclear reactor safety to inmate misconduct assessments, according to a report by authors who survey AI in democratic processes.
How is the US government using AI in federal agencies?
The federal government is shifting decision-making from humans to machines across public health, national security, and judicial functions. According to the OMB disclosure, the current inventory of AI applications has grown significantly, including several high-stakes implementations.
The Health and Human Services (HHS) office of administration for children and families has contracted Palantir to scan grant applications. The goal is to flag applications that don’t align with administration dictates. Meanwhile, the Federal Bureau of Prisons is building an AI system to predict “potential for misconduct” for new inmates, which could route individuals into high-security confinement before an incident occurs.
Other agencies are applying AI to life-and-death scenarios. The Department of Veterans Affairs is developing AI to monitor veterans crisis line calls and use external databases to assess suicide risk. In the energy sector, the Department of Energy is testing autonomous AI to control nuclear reactors and respond to safety incidents.
Why is the lack of transparency causing concern?
Critics argue the OMB’s disclosures lack the context needed to understand how these systems actually work. Most entries in the inventory are limited to a single sentence or a short paragraph, leaving the specific approach and purpose unclear.
Public consultation is largely absent from the process. Under current policy, agencies aren’t required to engage the public unless a use case is labeled “high impact.” However, the authors of the report note that this label is applied inconsistently across different agencies.
Currently, only the Department of Justice has proposed involving the public in its AI implementation. Most citizens only find these disclosures by monitoring the federal chief information officer’s GitHub account or reading specialized outlets like FedScoop.
How do US AI policies compare to France and Canada?
The US approach to AI disclosure is less rigorous than systems adopted in Europe and North America. While the US uses a list of use cases, other nations mandate legal protections and impact assessments.
| Country | Primary Mechanism | Key Requirement |
|---|---|---|
| United States | OMB Inventory | Disclosure of active/planned use cases. |
| France | Digital Republic Act (2016) | Right to human review and public records requests. |
| Canada | AI Use Case Registry | Mandatory risk-scoring and impact assessments. |
France’s 2016 law requires that any algorithm used for administrative decisions be appealable to a human reviewer. Canada’s system requires a detailed explanation of risks and benefits, along with stakeholder consultation, before a system is even conceived.
What are the beneficial applications of government AI?
Despite the risks, some AI deployments provide immediate utility. Machine translation is a primary example. Customs and Border Protection (CBP) uses AI translation to assist officers when human interpreters aren’t available.
The OMB inventory shows that translation use cases have increased from 58 under the Biden administration to 70 under the Trump administration. While human interpreters are better at reading social cues, an AI translator provides a baseline of communication that prevents total silence between officers and non-English speakers.
What happens next for AI regulation in government?
The trend is moving toward a demand for “algorithmic impact risk assessments.” Experts suggest the US should adopt federal and state-level procedures that require public comment periods before sensitive AI systems are deployed.
The goal is to transition from simple disclosure to active transparency. This would involve requiring agencies to respond substantively to public feedback, similar to the Canadian model, to ensure that AI improves efficiency without sacrificing equity or human rights.
Frequently Asked Questions
How many AI use cases are currently in the US government?
According to the OMB, there are 3,611 active or planned AI use cases across federal agencies.
Which companies are providing AI to the US government?
The report specifically names Palantir as a provider for the Health and Human Services (HHS) office of administration for children and families.
Is the public consulted on government AI use?
Generally, no. Public consultation is only required for “high impact” use cases, a label the report claims is applied inconsistently.
Do you think the US should adopt a human-review requirement for all AI-driven government decisions? Let us know in the comments or subscribe to our newsletter for more updates on AI policy.