“Representatives Sara Jacobs and Don Beyer are drafting legislation that would establish minimum safety standards for AI laboratories, clarify liability rules for AI-related harms, and grant the government emergency authority to take down models that fail safety testing. The bill represents one of the more concrete federal legislative attempts to regulate AI at the development stage. If passed, it could fundamentally reshape how AI labs operate, test, and release models in the United States.”
Key Takeaways
- Reps. Sara Jacobs and Don Beyer are preparing a federal AI safety bill targeting AI laboratories directly.
- The bill would establish minimum safety standards and clarify liability rules for AI developers.
- Government would gain emergency authority to shut down AI models that fail safety evaluations.
A new bill could give the government power to shut down dangerous AI models.
trending_upWhy It Matters
If enacted, this legislation would mark a significant shift from voluntary AI safety commitments — which have dominated the US approach so far — toward binding federal regulation. AI labs like OpenAI, Anthropic, and Google DeepMind could face mandatory testing regimes and legal liability exposure they currently avoid. The emergency shutdown provision is particularly consequential, as it would give regulators direct leverage over model deployment, not just guidelines. Lawmakers and lobbyists will be watching closely to see whether this bill gains bipartisan traction or stalls in a divided Congress.
FAQ
What safety standards would AI labs have to meet under this bill?
The bill proposes minimum safety standards, though specific technical thresholds have not yet been detailed publicly. Labs would likely need to demonstrate that models pass defined safety evaluations before or during deployment.
Who would have the authority to shut down an AI model, and how would that work?
The bill would grant the government emergency authority to shut down models that fail safety testing, though the specific agency responsible has not been confirmed. This would be a significant new regulatory power with no clear existing precedent in US tech law.
How does this bill differ from existing US AI policy efforts?
Most existing US AI policy, including the Biden-era executive order, relies on voluntary commitments and guidance rather than enforceable standards. This bill would create binding legal obligations and explicit liability rules, moving the US closer to the regulatory approach seen in the EU AI Act.
