Over the past few months, it has become clear that the absence of robust federal AI safety rules in the U.S. is unsustainable and unacceptable, given the acceleration in AI development and the rise in AI cyberattacks.
But this will likely change soon.
With Democrats having taken control of both the House and the Senate in the United States, Americans should expect new federal AI laws in 2027, particularly on AI safety, as various Democratic lawmakers have already signaled their intention to regulate AI, marking a sharp shift from the Republican-led Congress.
Given the urgency of AI security and safety challenges and the U.S. political shift, I predict the U.S. will soon regulate AI much more thoroughly and strictly than the EU, at both the state and federal levels.
Every AI governance professional should pay attention to this shift, which I cover in detail in my AI Governance Leadership Program and often discuss in our AI Governance Workshops. If you work in AI governance, I recommend you join both.
Some of the proposed AI safety frameworks look promising, and they make me much more hopeful about the future of AI and our ability to govern it effectively.
For example, a new bill proposed yesterday, the SAFE AI Act (Safe and Accountable Federal Evaluations of AI), would mandate safety frameworks, incident reporting, and increased transparency from frontier labs, with strong government oversight.
According to the preliminary outline published yesterday, these would be the main topics covered by the bill:
Developers publish enforceable safety frameworks.
Minimum government requirements.
Increased transparency on Automated AI R&D in accordance with prescribed measures.
Flexibility to evolve: Ability for scope of coverage, risk domains, and requirements to evolve over time.
Incident reporting and disclosures.
Government teams evaluate each developer.
Third-party verifiers accredited by the government to conduct compliance evaluations in their accredited domains.
Fee-based funds: Developers do not pay evaluators directly.
True independence of evaluators.
Pre-deployment evaluations: Evaluations assess framework compliance before commercial release or broad internal use.
Provides narrow emergency authority with judicial review: Where safeguards are inadequate and an AI system poses imminent risks of severe harm, the government may order mitigation measures.
Civil penalties for violations of safety frameworks, with opportunities for remediation and judicial review, and enhanced penalties for knowing or willful violations.
Clarifies liability of developers for reasonably foreseeable injuries to third parties caused by AI systems.
Whistleblower protection.
These developments make me hopeful for the future of AI and AI governance.
I look forward to discussing these new bills with you and seeing them implemented in practice. Stay tuned!




I hope the drafters have an eye towards how to avoid Major Questions Doctrine vulnerability. Given the rapid and unpredictable changes in the technology, it's not hard to envision this new regulator having to confront circumstances that were not explicitly contemplated in the legislation or by its drafters. Hmmm.