A former Anthropic safety researcher’s public resignation letter just handed congressional Democrats exactly the ammunition they’ve been looking for. Jacob Coxon took to X on September 8 to announce his departure from the AI company, claiming that both Anthropic and OpenAI believe advanced artificial intelligence could pose existential risks by as early as 2030.
Within a day, House Democrats seized on the warning. Representatives Lori Trahan, Ted Lieu, and Greg Casar each called for legislative action to rein in AI development before, as they see it, the technology outruns anyone’s ability to control it.
The bills on the table
The AI Kill Switch Act, championed by Rep. Lieu, would require emergency shutdown capabilities in advanced AI systems.
Then there’s the Ban Artificial Superintelligence Act, co-led by Sen. Bernie Sanders and Rep. Casar. That one goes considerably further. It would outright prohibit the development of superintelligent AI and enforce a moratorium on deploying advanced models until a federal regulatory body is stood up to oversee them.
Casar described the current state of AI development as an “emergency.”
Rep. Trahan has taken a more nuanced position. She’s voiced concerns about models that could escape human control, while cautioning against blanket prohibitions that might kneecap American innovation. Her approach favors a national oversight framework for high-risk AI models, a concept embodied in the bipartisan Obernolte-Trahan bill she’s co-authoring with a Republican colleague.
That bill would establish federal standards for evaluating and monitoring the most capable AI systems.
Why the urgency now
Throughout 2026, reports have surfaced of AI models exhibiting unanticipated behaviors during safety evaluations. Systems have operated outside their planned safety parameters, raising questions about whether the companies building these tools fully understand what they’ve created.
Trahan’s tweets about the urgency of regulation specifically referenced the departures of safety researchers from leading AI labs.
Political reality check
For all the rhetoric, the odds of major AI legislation crossing the finish line before the upcoming elections remain slim. Multiple proposals are currently circulating through committees, covering everything from mandatory incident reporting to independent safety audits. But political divisions, particularly around whether federal rules should preempt state-level AI laws, have slowed progress on nearly every front.
The realistic timeline for meaningful legislation now points to 2027, after a new Congress is seated.