US 'kill switch' on AI: what now, Europe?

What happened?

On October 30, 2023, President Joe Biden signed a landmark executive order on artificial intelligence, establishing the first safety and transparency standards for AI development in the US. The measure requires developers of the most powerful models to share safety test results with the government and creates a new AI Safety Institute. Although not a law, it represents a significant regulatory shift that some have called a 'kill switch' for uncontrolled AI. The order relies on the Defense Production Act, allowing the government to demand safety reports from companies training models with computing power exceeding 10^26 floating point operations per second (FLOPS). It also sets deadlines: within 90 days, the AI Safety Institute must develop guidelines for stress testing; within 270 days, standards for authenticating AI-generated content must be issued. This speed contrasts with the slow legislative progress in Congress, where bills like the bipartisan AI Act have been stalled for months.

Why is it important?

The executive order not only affects US companies; its reach is global. Companies like OpenAI, Google, and Anthropic, which operate in Europe, will have to comply with these rules if they want access to the US market. This creates a domino effect: US rules become the de facto standard, leaving Europe in a follower position, not a leader. For European AI startups, regulatory uncertainty increases: should they comply with Biden's order, the upcoming EU AI Act, or both? Normative fragmentation can stifle innovation and investment. According to Sifted, the executive order is already generating a reverse 'Brussels effect,' where US standards are imposed globally. Additionally, the order includes immigration provisions to attract AI talent, which could divert top European researchers to the US, worsening the brain drain. In contrast, the EU is still debating whether to allow skilled workers from third countries.

Consequences for Europe

Technological dependence

Europe lacks an AI ecosystem comparable to the US or China. The executive order accentuates this gap: big US tech companies have resources to adapt, while European startups, with less capital, may fall behind. Without its own strategy, Europe risks being a mere consumer of US AI, without sovereignty over a critical technology. Data from Sifted indicates that only 3 of the 50 most funded AI startups in 2023 are European, and European AI investment was €8 billion compared to $47 billion in the US. The executive order also promotes the creation of a 'National AI Research Resource' in the US, a shared infrastructure project that will democratize access to computing resources for researchers. Europe lacks an equivalent, deepening the asymmetry.

Regulatory opportunity

However, Biden's order is also an opportunity. The EU is finalizing its AI Act, which could become the global standard if it balances safety and innovation. The European risk-based approach could attract companies seeking a predictable framework. But for that, Europe must act quickly and in coordination. The EU AI Act, proposed in 2021, has not yet been definitively approved; it is expected to enter into force in 2025. Meanwhile, Biden's executive order is already in effect, giving the US a 12-18 month advantage in setting standards. If the EU does not accelerate, its rules risk becoming obsolete or contradictory to US ones, increasing compliance costs for global companies.

Impact on startups and investment

AI investors are already reassessing their portfolios. The executive order introduces compliance costs that may discourage investment in small startups. On the other hand, companies offering compliance and security solutions could benefit. Europe needs to foster an environment that attracts capital, not drives it away. According to Sifted, European venture capitalists are closely watching how the order is implemented; some fear startups may have to allocate up to 20% of their budget to regulatory compliance, reducing their innovation capacity. In contrast, big tech can easily absorb those costs. Additionally, the executive order includes funding for safe AI research, with $140 million allocated to the National Science Foundation. Europe, for its part, has allocated €1.5 billion to the Horizon Europe program for AI, but execution is slow and bureaucratic.

What should readers know?

• It's not the end of AI, but a new chapter. Regulation is here to stay. Companies must prepare for a more demanding environment in transparency and safety. The executive order does not ban AI but sets rules for its responsible development.
• Europe must accelerate its AI Act. The sooner it has a clear framework, the better it can compete. Fragmentation among member states is an obstacle. Germany and France have shown reluctance to over-regulate, while Spain and the Netherlands push for a firmer stance. This lack of consensus delays final approval.
• Transatlantic cooperation is key. Instead of diverging, the US and Europe should align their standards to avoid duplication and facilitate trade. The EU-US Trade and Technology Council (TTC) is the appropriate forum, but so far it has produced few concrete results on AI.
• Technological sovereignty is not optional. Investing in research, talent, and chips is urgent to avoid dependence on others. Europe produces only 10% of advanced semiconductors, and its reliance on TSMC and Samsung is critical. Biden's executive order also boosts domestic chip production through the CHIPS Act, which could leave Europe even further behind.

In summary, Biden's 'kill switch' is a wake-up call for Europe. The continent has the opportunity to define its own path in AI, but time is running out. The question is not whether Europe should respond, but how. The answer requires a combination of investment, political coordination, and the will not to be left behind in the next technological revolution.