AI Safety, Recess Edition

Another routine existential crisis solved by more recess

There's a semi-regular tradition in American politics for a news story materializing in the late summer of even-numbered years that dominates attention and causes a great deal of anxiety throughout the fall before dissipating into our memories around the first Wednesday after the first Monday in November. You may remember bravely surviving the Ebola threat of 2014. Or breathing a sigh of relief that all it took to repel the marauding migrant caravan from our borders was for cable TV stations to stop talking about them. You could be forgiven for thinking that this year's iteration—AI robots enslaving and/or murdering all of us—borrowed a suspiciously large number of plot elements from old Arnold Schwarzenegger movies and will similarly be forgotten by Thanksgiving until its sequel is announced in two years. This year's existential panic could be different though as it's being driven not by overzealous cable TV news hosts, but by some of the very people who are building the AI technology that they purport to not be able to control. Earlier this month, one former employee of Anthropic posted on social media that "the people building AI earnestly believe that it could kill us all by the end of the decade." Another employee posted that he and his co-workers believed that the very technology they are working on could kill us all and put the odds of that happening at over 10%. The CEO of Anthropic published an essay arguing that AI development needed to be "paced" while rival OpenAI announced that its planned IPO would be delayed due to these concerns. One hundred and seven House Democrats demanded Speaker Mike Johnson respond by urgently passing binding AI safeguards. And the Speaker responded…by canceling the last day of session and starting the pre-election recess he had already announced would be starting two weeks early. He echoes the dismissiveness of President Trump, who has described the safety warnings from frontier labs and their employees as a "hoax" and proclaimed that the only needed guardrail is a "STRONG AND SMART (High IQ!) PRESIDENT." The House is not scheduled to return until after November 3, so, assuming that the threat of AI bringing about human extinction hasn't either passed or tragically been realized, AI legislation will now have to wait until a lame-duck session — or, more likely, until the next Congress when Democrats are broadly expected to control at least the House and possibly even the Senate.

The House has ideas, but no time

This week's House Democrat letter called on Johnson to bring the House back "immediately and remain in session until Congress advances meaningful, bipartisan AI safeguards" and pointed to two bipartisan bills as the starting point. The first is the FRONTIER Act, introduced July 23 by Trahan and Rep. Jay Obernolte (R-CA), along with Reps. Scott Franklin (R-FL), Scott Peters (D-CA), Erin Houchin (R-IN), and Suhas Subramanyam (D-VA). The bill establishes a national, risk-based framework for the most advanced AI models, with tiered requirements scaled to developer size: model cards, risk-management frameworks, independent audits, and incident reporting. It also gives the Secretary of Commerce emergency authority to suspend development or deployment of a model that presents an imminent catastrophic risk. A controversial provision would preempt new state AI laws on auditing, incident reporting, and risk disclosures for three years — a feature that drew opposition from more than 200 state legislators across 42 states when an earlier draft circulated. The second bill is the AI Kill Switch Act, introduced the same day by Rep. Ted Lieu (D-CA) and Rep. Nathaniel Moran (R-TX). It would require frontier AI developers to build in technical shutdown controls and allow the Department of Homeland Security to compel model suspensions during emergencies. Notably absent from their requests was an outright moratorium on development of the frontier models themselves. The 107-member letter represents a majority of the 213 House Democrats, but the caucus continues to debate the issue. Progressive Caucus chair Rep. Greg Casar (D-TX) was joined by Sen. Bernie Sanders (I-VT) to announce the Ban Artificial Superintelligence Act to permanently ban the development and deployment of superintelligent AI and pause advanced AI development until a new cabinet-level federal agency establishes safety rules. Companies that violate the restrictions could face what Sanders described as a "corporate death penalty." Alternatively, the moderate Problem Solvers Caucus's AI task force, led by Reps. Josh Gottheimer (D-NY) and Mike Lawler (R-NY), took a narrower approach, introducing the bipartisan Stop Rogue AI Act on September 9 — focused on AI agents rather than frontier models broadly.

The Senate has time, but not for AI

The Senate remains in session for another two weeks after the House left (or, perhaps more accurately, it has not yet announced an early start to its planned pre-election recess) and conversations are ongoing. On September 16, Sen. John Kennedy (R-LA) tried to fast-track his AI Emergency Button Act via unanimous consent — a Senate version of the kill-switch concept — and was blocked by Sen. Rand Paul (R-KY), who argued that an open-ended mandate could affect AI systems throughout the economy. Meanwhile, the Senate Commerce Committee continues to discuss a proposal led by Sens. Ted Cruz (R-TX) and Amy Klobuchar (D-MN) that targets catastrophic AI risks by imposing a "legal duty of care" on frontier AI developers and creates a mechanism for the federal government to block unsafe model releases before they reach the public. Sen. Cruz's Democratic counterpart on the Commerce Committee, Sen. Maria Cantwell (D-WA), has objected to the proposal in its current form as she would prefer the models be tested by government scientists and experts as opposed to the developers themselves and has reservations about the bill's preemption of more stringent state laws. As these conversations percolate through the Senate, leadership has announced a focus on issues like college sports regulations and executive branch nominations for its floor proceedings.

After Election Day outlook…

Elections have a way of scrambling the leverage of the parties in Congress and this year could be exceptionally so. While both Houses are scheduled to return after Election Day in November, it is difficult to anticipate how the Republican majority would spend its waning days in power should Democrats regain control of either body. While there could be pressure to use their final legislative breath on partisan bills such as another reconciliation package or completing work on annual appropriations and routine authorization bills, the sheer number of departures from Congress implies that it will be difficult for the already narrow majority to force a robust agreement on anything through on their own. Barring additional urgency that forces bipartisan agreement for AI legislation—in other words, something very bad happens—it is more likely that Congress does what it does best and punts the issue into 2027, where a new Congress awaits.