|
Brussels, |
|
Following Europe's Lead
U.S. Congress to Regulate A.I., Industry Pushes Back
Congress proposes emergency powers over frontier AI as the White House and Big Tech's Open Weights coalition debate America's regulatory future.
A bipartisan proposal in Congress would grant the US government emergency powers to throttle or shut down advanced AI systems capable of causing catastrophic harm, bringing Washington closer to the risk-governance philosophy pioneered by the European Union. At the same time, the White House is pursuing a more targeted national-security approach, while the Open Weights coalition—led by Nvidia, Microsoft, Meta, OpenAI, Google and dozens of other technology companies—argues that excessive restrictions on open-weight AI models could undermine innovation, competition and American leadership. The emerging debate is no longer simply about regulating artificial intelligence. It is about who should control frontier AI, how open its most powerful models should remain, and how to balance public safety with technological progress.
By Paolo Licandro
Brussels, 28 July, 2026
For years, the United States contrasted its innovation-led approach to artificial intelligence with the European Union’s regulatory model. That distinction is now becoming less clear.
Democratic Representative Ted Lieu and Republican Representative Nathaniel Moran have introduced the bipartisan AI Kill Switch Act, which would require developers of exceptionally powerful AI systems to retain the technical ability to throttle, suspend or shut them down. The Department of Homeland Security could order emergency intervention when a system presents a credible risk of catastrophic harm, including loss of human control, threats to critical infrastructure or severe economic disruption.
The proposal reflects a principle already embedded in the European Union’s Digital Policy framework: advanced AI cannot be treated as an ordinary commercial product when its capabilities may create systemic risks.
Through the AI Act, the EU has already established legally binding obligations based on risk, human oversight, transparency and accountability. The American bill is narrower and more emergency-oriented, but it follows the same fundamental logic: governments must retain enforceable powers when voluntary safeguards are insufficient.
Congress: Control Must Remain Technically Possible
The congressional sponsors argue that an emergency power is useless unless developers have designed their systems to obey it. Their proposal therefore focuses on technical controllability.
This position has gained urgency following reports of advanced AI systems behaving unexpectedly during cybersecurity testing. Supporters maintain that catastrophic-risk legislation should be established before—not after—a model escapes meaningful supervision.
Their case is essentially precautionary: society does not wait for an aircraft accident before requiring emergency systems, nor should it wait for an uncontrollable AI incident before requiring a shutdown mechanism.
The White House: Innovation Under National-Security Supervision
The White House occupies a more nuanced position. Its June 2026 executive order on advanced AI promotes rapid deployment and American technological leadership, while also creating coordinated federal evaluation of frontier models’ cybersecurity capabilities.
The administration has avoided endorsing a general licensing regime for AI releases. Nevertheless, it has demonstrated a willingness to intervene selectively where frontier models, foreign access, cybersecurity or alleged technological appropriation raise national-security concerns.
Its approach is therefore neither European-style horizontal regulation nor complete industry self-governance. It favours federal oversight concentrated on strategic capabilities, trusted access and threats from foreign competitors—particularly China.
The Industry Letter: Openness Is Part of the Defence
A different argument appears in the letter “Open Weights and American AI Leadership,” initially signed by Nvidia, Microsoft, Meta, IBM, Palantir, Mistral, Hugging Face and other technology and investment organisations.
The coalition warns against “premature restrictions” on open-weight models—systems whose trained parameters can be downloaded, examined, adapted and operated independently.
Its central claim is that openness is not necessarily the enemy of safety. Open models allow researchers to identify vulnerabilities, companies to create specialised safeguards and users to avoid dependence on a handful of closed platforms. They also reduce costs, strengthen competition and enable universities, start-ups and public authorities to develop applications without paying frontier-model prices.
The signatories argue that US leadership will depend not only on possessing the most powerful proprietary model, but on spreading AI capabilities throughout the economy. Broad restrictions, they warn, could concentrate power, discourage innovation and push developers toward less restrictive jurisdictions.
OpenAI was absent from the initial signatory list but reportedly joined the expanded coalition later. Anthropic remains outside it. That distinction illustrates a deeper industrial divide: open-weight advocates prioritise diffusion and competition, while frontier-model developers also seek stronger protection against unauthorised distillation, misuse and replication of proprietary capabilities.
Three Visions of American AI Leadership
The disagreement is therefore not a simple choice between regulation and deregulation.
Congress emphasises emergency control. The White House prioritises national security and strategic technological leadership. The industry coalition defends openness, competition and widespread adoption.
Europe has already decided that enforceable safeguards are compatible with technological development, although debate continues over whether its rules may slow investment. The United States is now confronting the same fundamental question through a more fragmented political process.
The ultimate challenge is to combine all three objectives: preserving innovation, preventing excessive concentration and ensuring that the most powerful AI systems remain subject to human authority.
The transatlantic debate is converging on one conclusion: the future of AI will not be determined only by who builds the strongest model, but by who establishes the most credible rules for controlling, opening and governing it.
Brussels, 28 July, 2026
For years, the United States contrasted its innovation-led approach to artificial intelligence with the European Union’s regulatory model. That distinction is now becoming less clear.
Democratic Representative Ted Lieu and Republican Representative Nathaniel Moran have introduced the bipartisan AI Kill Switch Act, which would require developers of exceptionally powerful AI systems to retain the technical ability to throttle, suspend or shut them down. The Department of Homeland Security could order emergency intervention when a system presents a credible risk of catastrophic harm, including loss of human control, threats to critical infrastructure or severe economic disruption.
The proposal reflects a principle already embedded in the European Union’s Digital Policy framework: advanced AI cannot be treated as an ordinary commercial product when its capabilities may create systemic risks.
Through the AI Act, the EU has already established legally binding obligations based on risk, human oversight, transparency and accountability. The American bill is narrower and more emergency-oriented, but it follows the same fundamental logic: governments must retain enforceable powers when voluntary safeguards are insufficient.
Congress: Control Must Remain Technically Possible
The congressional sponsors argue that an emergency power is useless unless developers have designed their systems to obey it. Their proposal therefore focuses on technical controllability.
This position has gained urgency following reports of advanced AI systems behaving unexpectedly during cybersecurity testing. Supporters maintain that catastrophic-risk legislation should be established before—not after—a model escapes meaningful supervision.
Their case is essentially precautionary: society does not wait for an aircraft accident before requiring emergency systems, nor should it wait for an uncontrollable AI incident before requiring a shutdown mechanism.
The White House: Innovation Under National-Security Supervision
The White House occupies a more nuanced position. Its June 2026 executive order on advanced AI promotes rapid deployment and American technological leadership, while also creating coordinated federal evaluation of frontier models’ cybersecurity capabilities.
The administration has avoided endorsing a general licensing regime for AI releases. Nevertheless, it has demonstrated a willingness to intervene selectively where frontier models, foreign access, cybersecurity or alleged technological appropriation raise national-security concerns.
Its approach is therefore neither European-style horizontal regulation nor complete industry self-governance. It favours federal oversight concentrated on strategic capabilities, trusted access and threats from foreign competitors—particularly China.
The Industry Letter: Openness Is Part of the Defence
A different argument appears in the letter “Open Weights and American AI Leadership,” initially signed by Nvidia, Microsoft, Meta, IBM, Palantir, Mistral, Hugging Face and other technology and investment organisations.
The coalition warns against “premature restrictions” on open-weight models—systems whose trained parameters can be downloaded, examined, adapted and operated independently.
Its central claim is that openness is not necessarily the enemy of safety. Open models allow researchers to identify vulnerabilities, companies to create specialised safeguards and users to avoid dependence on a handful of closed platforms. They also reduce costs, strengthen competition and enable universities, start-ups and public authorities to develop applications without paying frontier-model prices.
The signatories argue that US leadership will depend not only on possessing the most powerful proprietary model, but on spreading AI capabilities throughout the economy. Broad restrictions, they warn, could concentrate power, discourage innovation and push developers toward less restrictive jurisdictions.
OpenAI was absent from the initial signatory list but reportedly joined the expanded coalition later. Anthropic remains outside it. That distinction illustrates a deeper industrial divide: open-weight advocates prioritise diffusion and competition, while frontier-model developers also seek stronger protection against unauthorised distillation, misuse and replication of proprietary capabilities.
Three Visions of American AI Leadership
The disagreement is therefore not a simple choice between regulation and deregulation.
Congress emphasises emergency control. The White House prioritises national security and strategic technological leadership. The industry coalition defends openness, competition and widespread adoption.
Europe has already decided that enforceable safeguards are compatible with technological development, although debate continues over whether its rules may slow investment. The United States is now confronting the same fundamental question through a more fragmented political process.
The ultimate challenge is to combine all three objectives: preserving innovation, preventing excessive concentration and ensuring that the most powerful AI systems remain subject to human authority.
The transatlantic debate is converging on one conclusion: the future of AI will not be determined only by who builds the strongest model, but by who establishes the most credible rules for controlling, opening and governing it.
