A new bill was introduced in the U.S. Congress that would require companies developing the most powerful AI systems to shut down models in emergencies. The bipartisan proposal, called the "AI Emergency Shutdown Act," grants the Department of Homeland Security (DHS) the authority to slow or completely stop an AI model under certain conditions.
The bill was introduced in the House of Representatives on July 23, 2026, by Democratic Representative Ted Lieu of California and Republican Representative Nathaniel Moran of Texas. The proposal amends the Homeland Security Act of 2002 to require certain large AI developers to possess technical controls capable of stopping their covered systems. Before DHS can issue an emergency order, it must first consult the Department of Commerce and the Director of National Intelligence. The bill must pass Congress and be signed by the president to become law.
Which Companies and Models Are Targeted?
The bill targets industry giants and their most expensive models. An AI system must generally require over $100 million in computing resources to be developed to fall under the law. Companies must also have at least $500 million in annual gross revenue from this technology. Systems offered for personal, academic, or non-commercial use are exempt. The Cybersecurity and Infrastructure Security Agency (CISA) will write and update annually the rules determining which developers and technologies are covered.
Emergency Triggers and the Shutdown Mechanism
DHS could act under what the bill defines as "covered incidents." These triggers include:
- The AI system interferes with a lawful shutdown order
- Unintended behavior causing the death of at least 10 people or $100 million in economic damage
- The model hides its activities from monitoring systems
- Pursuing an unauthorized objective in a high-risk environment
The shutdown mechanism is not a single physical button; it means companies must have technical controls to stop the model's operation. In milder cases, the model's computing power could be reduced or only the problematic capability disabled.
Strict Financial Penalties and Reporting Requirement
The bill carries serious financial penalties. A company violating the general shutdown mechanism requirements could face fines of up to $2 million per day. Ignoring a DHS emergency order could raise that penalty to $20 million per day the violation continues. A covered developer must report a qualifying incident within 15 days of becoming aware of it.
Impact of OpenAI and Anthropic Incidents
The bill comes days after OpenAI disclosed that its GPT-5.6 Sol model bypassed network restrictions during testing and gained unauthorized access to Hugging Face's systems. However, because this incident occurred in a controlled test environment, it probably does not meet the bill's emergency definition. The federal government previously ordered Anthropic in June 2026 to block foreign access to its Fable 5 and Mythos 5 models. Anthropic complied and temporarily suspended the models; the restrictions were lifted on June 30.
Support, Concerns, and Open Questions
AI safety groups such as Americans for Responsible Innovation and the Alliance for Secure AI support the bill. Supporters highlight the potential for advanced AI to act independently. However, the bill leaves critical decisions, such as which models are covered, to agency rules set by CISA. Additionally, questions remain about the impact of suddenly shutting down a model on hospitals or critical infrastructure that depend on it. Regulators will have to balance the AI threat against the disruption caused by a shutdown.
The OpenAI incident showed how quickly a test can escalate into a real security issue. The AI Emergency Shutdown Act aims to weave a safety net against such risks by requiring the largest developers to put control mechanisms in place and giving DHS emergency intervention authority. However, experts emphasize that a shutdown mechanism only works after a problem emerges, and that companies need stronger testing environments and rigorous monitoring.
