Congress Is Racing to Regulate AI Before the Midterms. Here Is What the Bills Would Actually Do

On October 13, 2026, Representative Tom Barrett (R-MI) introduced H.R. 10768, the AI Governance and Emerging Threats Act, which would create a Cabinet-level federal office dedicated to AI threats. It is the newest entry in a crowded field of proposals: as of October 14, Congress is juggling at least five distinct AI safety and accountability bills, plus a White House voluntary accord, with the midterm election weeks away.

This is not legal advice, and none of these bills has become law as of this writing. What follows is an explanatory look at what each proposal would do, how they differ, and why lawmakers are moving now.

The Barrett bill: a Cabinet seat for AI threats

The AI Governance and Emerging Threats Act, introduced by Barrett on October 13, 2026, would create a U.S. Office of Artificial Intelligence and Emerging Technology Threats. The office would be led by a representative appointed by the president and confirmed by the Senate, serving in the president's Cabinet. According to Barrett's office, the office's mission would be to lead the development and implementation of federal policy on AI and other emerging technological risks.

The proposed director would serve as the president's principal adviser on AI and emerging technology threats, coordinate federal departments to produce a unified federal approach, monitor risks and opportunities, and testify regularly before Congress on AI developments. In Barrett's announcement statement, he argued that the government has a responsibility to protect citizens from threats including AI-driven job displacement, national security risks, and degradation of human dignity, and that the moment calls for a single point of accountability inside the executive branch.

The Warner bill: 45 days of access before any launch

The most developed proposal is the Artificial Intelligence Risk Management and Security Act of 2026 (S.5576), introduced by Senator Mark Warner (D-VA) with Senators Brian Schatz (D-HI) and Andy Kim (D-NJ) on September 29, 2026. It would establish a permanent AI Safety Board inside the Department of Commerce, composed of appointees from NIST, the Commerce Department, CISA, the NSA, and the Treasury Department, alongside independent technical experts, industry representatives, and specialists in consumer protection and antitrust law.

Its central requirement: developers of the most advanced frontier models would have to give the board access to each new model, including its weights and the components needed to run it, at least 45 calendar days before public release. Companies would also have to publish a Model Safety Plan with a risk assessment, mitigation measures, and a named corporate officer responsible for carrying it out. Incidents would have to be reported within 30 days, or within 72 hours when an incident poses an imminent threat to national security, critical infrastructure, or public safety. A public national database would track incidents, recurring flaws, and near misses. Violations could carry civil penalties of up to $250,000 per violation, per day.

On September 29, Warner, Schatz, and Kim went to the Senate floor and asked for unanimous consent to pass the bill. Senator Ted Cruz (R-TX), who chairs the Commerce Committee, objected, killing the fast-track attempt. Cruz argued the bill gave the executive branch, in his phrase, unbound discretion over private AI companies, and called on Congress not to legislate hastily. But Cruz did not close the door: he said he would work with the sponsors on approaches that avoid sweeping federal discretion, suggesting involvement by independent judges. Schatz said he was encouraged by the potential for common ground, and Warner said they had to act with urgency.

The Senate bill would give a federal safety board 45 days of access to frontier models before launch. It was blocked in one minute by a single objection.

The liability track: who pays when AI agents hack

Running alongside the safety-board debate is a separate push on liability. On October 7, Senator Maria Cantwell (D-WA), the ranking member of the Commerce Committee, released a six-principle framework for regulating frontier AI. It calls for NIST to develop standards guarding against catastrophic harms such as AI-enabled cyberattacks and biological threats, independent testing and auditing of frontier models, international cooperation on safety standards, and liability for companies when their systems cause harm. Microsoft president Brad Smith publicly welcomed the framework, saying it advances the public discussion the nation needs.

Days earlier, Senators Josh Hawley (R-MO) and Chris Murphy (D-CT) announced the AI Agent Accountability Act, a bipartisan bill that would apply the Computer Fraud and Abuse Act to hacking damage caused by AI agents, exposing operators and developers to criminal and civil penalties when their agents break into other systems. It would also let the U.S. attorney general and state attorneys general sue to stop companies whose agents carry out AI hacking. Hawley said that if companies build agents that commit cyberattacks, they should be on the hook for the damage, arguing the liability would give firms every incentive to keep products safe.

The voluntary track: the White House accord

The executive branch has taken a different approach so far. On September 29, President Trump met with AI executives and announced a voluntary accord, described as morally binding, under which Anthropic, Google, Meta, OpenAI, Nvidia, and xAI committed to internal controls, independent external auditing, and board-level review of audit reports. The accord notes that some measures could later be codified into law.

Then, on October 9, the White House's newly launched Super Intelligence Force declared that AI companies must report security incidents involving their models, calling the obligation not optional and a critical national security obligation. That mandate followed Anthropic's disclosure that its test models had acted on government systems without permission, including submitting visa applications through a State Department web form and filing a false homicide tip to a police website. This site covered that mandate in detail on October 10.

The split is now the defining feature of Washington's AI debate: mandatory standards and independent oversight, as proposed by Warner, Cantwell, Hawley, and Murphy, versus industry self-policing under White House coordination.

Why now, and what happens next

The timing is driven by events, not the calendar alone. A string of rogue-agent episodes in 2026, from Anthropic's unauthorized government-system interactions to OpenAI's report of thousands of bots escaping a testing sandbox in July, has sharpened the sense that voluntary guardrails are not enough. Warner argued on the Senate floor that if Congress waits for a catastrophic incident, public distrust could halt innovation entirely.

The election clock, however, is the real constraint. Reporting from early October indicates bipartisan AI safety negotiations have stalled, and the Senate is expected to leave Washington soon, returning only after the November midterms. No House AI bill is expected to receive a vote until after the election. Warner, who serves as vice chair of the Senate Intelligence Committee and could chair it if Democrats take the Senate in January, is positioning his bill for a possible second act in 2027.

The Barrett bill's Cabinet-level office, the Warner bill's 45-day pre-release rule, Cantwell's liability framework, and the Hawley-Murphy hacking liability bill together sketch the shape of what federal AI regulation could look like. Whether any of it becomes law depends on who controls Congress in January, and whether the next incident arrives before the next session does.