A wave of renewed urgency is building in Washington around AI safety legislation, and this time it isn’t coming from a single party or a single chamber. A bipartisan Senate bill is reportedly close to introduction, progressive lawmakers are drafting far more aggressive proposals, and even major AI developers are publicly asking Congress to act. The catalyst: a stark warning from a departing AI safety researcher that has jolted lawmakers into treating catastrophic AI risk as an active policy problem rather than a distant hypothetical.
The Spark: A Warning From a Departing AI Researcher
The current momentum traces back to a round of blunt public warnings from a resigning AI researcher, who said that people building frontier AI systems genuinely believe those systems could “kill us all by the end of the decade.” Congress wasn’t even back in session when the comments began circulating among Hill staffers and lawmakers, according to reporting from Semafor. That single statement appears to have accelerated conversations that had been moving slowly for months, pushing several competing legislative efforts toward the front of the queue.
The Bipartisan Senate Bill Taking Shape
Behind closed doors, a bipartisan Senate proposal from Sens. Amy Klobuchar (D-Minn.), Ted Cruz (R-Texas), and Majority Leader John Thune has emerged as the option with the clearest path to actually passing before the end of the year. Sources familiar with the effort say it could be formally introduced as early as next week. Klobuchar described the goal as building a bipartisan agreement around commonsense guardrails to address the risks posed by AI systems, adding that lawmakers need to act now rather than wait. Cruz has confirmed he’s working with Klobuchar and Thune on legislation specifically targeting catastrophic risks tied to biological and nuclear threats.
What the Klobuchar-Cruz-Thune Proposal Would Actually Do
While the full draft text hasn’t been made public, early signals suggest the bill is centered on frontier AI systems — the most advanced, largest-scale models — rather than AI broadly. The stated focus on biological and nuclear risk echoes concerns raised by AI safety researchers for years: that sufficiently advanced systems could meaningfully lower the barrier for bad actors to design weapons or cause mass harm. Both frontier AI labs and public advocacy groups are already reviewing and providing feedback on the unreleased draft to congressional staff, suggesting the bill is further along than a typical early-stage proposal.
Progressives Push for a Tougher Approach
Not everyone on Capitol Hill sees the bipartisan Senate bill as sufficient. Sen. Bernie Sanders (I-Vt.) is preparing to introduce legislation that would go significantly further, reportedly aiming to ban the development of superintelligent AI systems altogether and pause AI development more broadly. Sanders is also planning a bipartisan briefing with outside experts in the coming days, following a string of recent AI-related security incidents that have heightened concern among lawmakers.
Rep. Ro Khanna (D-Calif.) has laid out his own five-point framework for addressing AI harms, calling for mandatory certification of new AI models before release, liability and criminal penalties for companies whose systems cause serious harm, and the creation of an entirely new federal agency dedicated to AI oversight. Compared to the Klobuchar-Cruz-Thune approach, these proposals reflect a fundamentally more interventionist view of how aggressively government should constrain frontier AI development.
AI Companies Are Already Lobbying on the Text
Perhaps the most notable sign that this legislative push has real momentum is that AI companies themselves are publicly asking for regulation rather than resisting it. OpenAI’s chief global affairs officer, Chris Lehane, wrote in a recent blog post that the company wants to work directly with Congress on national AI safety rules, pointing to what he called a new chapter in AI capability as the reason urgency is warranted. Lehane argued that Congress should prioritize a bias toward meaningful action over policy perfection, and warned that the current window for passing serious AI legislation may not stay open much longer.
That kind of public posture from a leading AI lab is a meaningful shift from the more defensive lobbying stance the industry has taken in past legislative cycles, and it suggests frontier labs see a real chance that some version of federal AI safety law passes before Congress adjourns.
This Isn’t Congress’s First Attempt at AI Rules
The current push builds on — and in some ways competes with — earlier federal efforts that have had mixed traction. Sen. Mark Warner has already rolled out a broader legislative package addressing AI infrastructure, competition, worker impacts, and national security, including a bill that would create a NIST-led testing and verification process for AI models. Separately, House Energy and Commerce Committee members introduced draft legislation known as the Great American AI Act, aimed at establishing federal oversight of frontier AI development while preempting individual state AI laws — a preemption approach that has drawn pushback from states that have already passed their own frameworks.
States Have Already Moved Faster Than Washington
While Congress debates federal approaches, state legislatures have been considerably more active. California’s Transparency in Frontier Artificial Intelligence Act and New York’s Responsible AI Safety and Education Act both impose transparency and safety-reporting requirements on developers of large frontier models, and dozens of other state bills targeting AI chatbot safety specifically have passed across nearly 20 states this year alone. That patchwork of state-level rules is part of what’s driving federal lawmakers to act — both to establish consistent national standards and, in some proposals, to override state laws entirely in favor of a single federal framework.
Why the Debate Over AI Risk Is So Divisive
The disagreement in Washington isn’t just about how to regulate AI — it’s about how seriously to treat the underlying risk itself. Some researchers and lawmakers argue that advanced AI systems pose genuine catastrophic or existential risks that justify aggressive intervention, including outright bans on certain types of development. Others, including many within the AI industry, argue that while real risks exist — particularly around bioweapons design assistance or critical infrastructure vulnerabilities — the more urgent priorities are practical safeguards, transparency requirements, and liability frameworks rather than broad development pauses. That underlying disagreement about the nature and scale of the risk is a big part of why proposals currently range from narrowly scoped safety guardrails to calls for banning entire categories of AI research.
What Happens Next
With the Senate not yet back in full session, the coming weeks will determine whether the bipartisan Klobuchar-Cruz-Thune bill can actually reach introduction and gather enough support to move through committee. Sanders’ planned expert briefing and Khanna’s proposed framework suggest that even if the bipartisan bill advances, it will face pressure from lawmakers who see it as insufficient. Meanwhile, industry statements like Lehane’s signal that AI companies are treating this as a genuine inflection point rather than routine political theater — a sign that some form of federal AI safety legislation may be more likely to pass in the coming months than it has been at any point so far in this Congress.
