AI Emergency Button Act
- Sponsor
- Rep. Kean, Thomas H. [R-NJ-7]
- Committees
- Science, Space, and Technology Committee (primary)
- Last action
- Sep 24, 2026
Bottom line
The bill requires a human-controlled 'emergency button' for advanced AI systems to ensure human oversight and control over potentially dangerous AI.
What it actually does
This bill mandates that all entities developing or operating advanced artificial intelligence (AI) systems in the United States must incorporate a technical capability allowing a human operator to shut down the system. It defines 'artificial intelligence' by reference to existing law and tasks the Secretary of Homeland Security with promulgating regulations to ensure compliance within 90 days of enactment.
Proponents argue
Proponents argue that this bill is a crucial step in ensuring human safety and control over increasingly powerful AI systems. They contend that a mandatory shutdown mechanism provides a necessary safeguard against unforeseen behaviors, malfunctions, or malicious uses of AI, preventing potential harm and maintaining human agency in critical situations. This proactive measure is seen as essential for responsible AI development.
Opponents contend
Opponents might argue that this bill could stifle innovation in the rapidly evolving AI sector by imposing potentially burdensome and technically challenging requirements. They may contend that a universal 'shutdown button' is not feasible or effective for all types of advanced AI, particularly highly distributed or autonomous systems, and could create a false sense of security without addressing the underlying complexities of AI safety. Furthermore, they might raise concerns about the broad definition of 'advanced artificial intelligence' and the potential for overregulation.
The bill is very short and can be read and understood quickly, allowing ample time for evaluation.
Section 2(b)
Mandatory Human-Controlled Shutdown Mechanism
This provision requires any entity that develops or operates an 'advanced artificial intelligence system' within the United States to build in a technical feature that allows a human operator to shut down the system. This is intended to provide a direct means for human intervention and control, acting as a safety switch for AI systems.
Supporters argue
Supporters argue that this provision is fundamental for establishing a baseline of safety and accountability in AI. They believe that human oversight, enabled by a clear shutdown mechanism, is non-negotiable for systems that could have significant societal impact, preventing catastrophic failures or misuse.
Critics contend
Critics might contend that mandating a universal shutdown mechanism could be technically challenging or even impossible for certain advanced AI architectures, such as highly distributed or continuously learning systems. They may also argue it could lead to a false sense of security, as a simple 'button' might not address the root causes of complex AI failures or ethical dilemmas, and could be easily circumvented or rendered ineffective in real-world scenarios.
Tradeoffs
The provision balances the desire for immediate human control and safety with the practical challenges of implementing such a mechanism across diverse and rapidly evolving AI technologies. It also creates a tension between fostering innovation and imposing regulatory requirements that could increase development costs and complexity.
Defines 'artificial intelligence' by referencing section 5002 of the National Artificial Intelligence Initiative Act of 2020 (15 U.S.C. 9401).
Section 2(a)(1)
Why it matters:This is standard legislative practice to ensure consistency with existing federal definitions and avoid redundancy, rather than an attempt to obscure the definition.
Case for: Referencing an existing, established definition ensures legal consistency and leverages prior legislative work, providing clarity for entities already familiar with federal AI policy.
Case against: For a casual reader, requiring external lookup adds a layer of complexity, potentially obscuring the exact scope of AI systems covered by the bill without consulting additional legal texts.
Estimated impact: The precise scope of the bill's applicability hinges on this definition, which is broad, encompassing various AI technologies from machine learning to expert systems.
Defines 'covered entity' as 'any entity that develops or operates an advanced artificial intelligence system in the United States.'
Section 2(a)(2)
Why it matters:This phrasing is likely intended to give regulatory agencies flexibility in defining what constitutes 'advanced' AI as technology evolves, but it also means the immediate scope is not fully clear from the bill text alone.
Case for: Granting regulatory agencies the power to define 'advanced' allows for adaptability as AI technology progresses, ensuring the regulations remain relevant without requiring new legislation for every technological shift.
Case against: The lack of a specific definition for 'advanced' creates uncertainty for entities, making it difficult to determine if they are subject to the bill's requirements until regulations are promulgated. This could lead to ambiguity and potential delays in compliance.
Estimated impact: The interpretation of 'advanced' will significantly determine the number and type of entities affected by the bill, ranging from major tech companies to smaller startups utilizing sophisticated AI.