As a wave of newly elected officials prepares to assume office in January, policy experts are pressing for swift legislative action to ensure that government use of artificial intelligence is transparent, congressionally approved, and can be halted when necessary. Currently, state and federal agencies employ AI to influence decisions ranging from criminal investigations to military operations, often without explicit legislative authorization or public disclosure.
Bill Gates recently underscored the need for mandatory safeguards and monitoring of AI, but comprehensive regulation remains elusive. However, experts argue that lawmakers need not wait for broader AI rules to govern their own institutions. They can immediately impose oversight on government AI use while broader debates continue.
The Three-Pronged Standard
Jim Townsend, director of the nonpartisan Levin Center for Oversight and Democracy, and Patrick Carter, a former Minnesota state budget official, propose a standard for government AI: transparent, approved, and stoppable. Transparency means a public registry of all AI systems used in decisions affecting individuals. Approval requires that no AI system operate without legislative authorization or safety certification. Stoppability ensures a designated official is accountable and that AI-influenced decisions can be reversed.
Transparency is critical because many agencies do not fully track their AI tools, which may be purchased, acquired free, bundled into existing software, or operated by contractors. Even non-AI automated systems have caused harm—such as Michigan's unemployment system falsely accusing thousands of fraud—and AI's complexity amplifies risks. Public visibility is essential to prevent abuse and errors.
The authors recommend that agencies publish lists of AI uses and notify individuals when AI contributes to decisions about them. Agencies should bear the cost of correcting AI mistakes, not the affected citizens. Legislative oversight should ensure compliance.
Approval and Oversight
Specific legislation should authorize AI deployment, setting criteria for vetting capabilities, limiting scope, and requiring safety certification. The risk of malicious manipulation is real, as demonstrated by researchers who tricked AI reviewers with hidden text. Corrupt officials or cybercriminals could similarly exploit government AI. Therefore, only AI systems approved by both legislature and agencies, with staff access to logs and test cases, should be permitted.
Stoppability and Accountability
Flawed systems can operate for years undetected, as seen in Michigan's unemployment scandal. AI could either reduce or exacerbate such failures, depending on whether officials are responsible for stopping malfunctioning systems. The authors stress that a named official must answer for AI use and that decisions must be reversible.
Some states—Texas, Indiana, California, and Connecticut—have already ordered inventories of automated systems, and a pending Senate defense bill would require contractors to disclose AI details to the Pentagon. These steps show growing legislative interest, but experts argue for more ambitious action.
Lawmakers have both the authority and the duty to lead on this issue. Voters should demand that government AI be transparent, approved, and stoppable, before and after the elections. As AI safety rules could demand unprecedented government surveillance, the need for clear guardrails is urgent. With public confidence in the federal government at a new low, demonstrating accountability on AI could help rebuild trust.
Incoming lawmakers, many of whom campaigned on AI risks, have an opportunity to enact meaningful oversight. By requiring transparency, approval, and stoppability, they can ensure that government AI serves the public interest without overstepping democratic boundaries.
