Over the past few months, I have been closely tracking how the global powers are attempting to tame the frontier of artificial intelligence. If you step back from the daily headlines and look at the actual mechanics of governance unfolding on both sides of the Atlantic, an obvious contrast emerges, one that makes you question whether we in Europe are actually taking the right path.
In the United States, the approach to AI governance has been surprisingly empirical. Rather than rushing to draft sweeping, all-encompassing laws, the US has essentially adopted a model of “soft regulation” centered on capability gating. They aren’t trying to regulate every potential software script under the sun; instead, they are focusing their lens directly on frontier AI models, the next-generation systems possessing emergent capabilities that could pose genuine national security risks.
When you hear industry leaders like Anthropic’s Dario Amodei warn about the near-term possibility of AI autonomously assisting in the design of biological weapons, or when you observe the sheer offensive cybersecurity capabilities emerging in systems like Mythos, GPT-5.6, and beyond, the US strategy makes pragmatic sense. It’s a targeted gatekeeping process. If a model crosses a specific threshold of power, where it could be hijacked by bad actors or hostile nation-states, it faces direct oversight. It is an approach grounded in the technical reality of how Large Language Models (LLM) and diffusion systems actually evolve.
Meanwhile, across the pond, Europe took immense pride in being first to market with the comprehensive EU AI Act. But as I reflect on its real-world impact, I find myself asking a fundamental question: Has this proactive regulation actually made us safer, or has it simply created a mountain of compliance costs?
Consider the human element. The EU AI Act was designed to protect citizens, yet we continue to witness tragic reports of individual harm, for example, people suffering severe mental health crises, falling into psychologically manipulative traps set by AI companions, or becoming detached from reality inside synthetic loops. The proactive, heavy-handed regulatory framework didn’t prevent these nuanced, deeply human vulnerabilities from occurring.
What it did do, however, was establish a culture where the default reaction to innovation is the threat of massive financial penalties.
This brings me to a core conviction: What if the answer to safe AI isn’t in penalizing companies for non-compliance, but in rewarding them for good behavior?
Instead of relying almost exclusively on the stick, European policymakers should start leveraging the power of the carrot. AI development requires massive capital investment. If Europe wants to build an ecosystem that is both competitive and ethically sound, we need to align market incentives with public good.
Imagine a governance model built on three positive pillars:
1. Co-Funding Conscious Innovation
Rather than waiting to audit a finished product and hand out fines, European state funds should offer direct grants to startups and enterprises that embed AI compliance and safety by design right from day one. If a company commits to building transparent, human-centric AI solutions, the state should help co-fund that journey.
2. Rewarding Reusable Safety Infrastructure
Building proper guardrails (such as robust data governance mechanisms, bias mitigation tools, or privacy-preserving architectures) costs time and money. The highest levels of grant funding should go to companies creating reusable safety components. If a startup builds a state-of-the-art privacy layer that can be open-sourced or shared across the European ecosystem, that effort should be heavily subsidized. By funding shared infrastructure, we accelerate the entire market’s safety baseline.
3. Cumulative Tax Rebates for Resilient Enterprises
For established businesses deploying AI at scale, compliance shouldn’t just feel like a line-item expense: it should feel like a competitive advantage. Imagine a system where enterprises achieving verified compliance across AI ethics, GDPR, and cybersecurity receive tangible tax incentives, say, a 5% tax rebate. If a company demonstrates cumulative resilience across data privacy, cyber defense, and AI safety, that tax relief compounds. Suddenly, boardrooms aren’t asking “How do we dodge this fine?” but “How quickly can we qualify for the resilience rebate?”
To be clear, this is not a call for total deregulation. The hammer must still exist, but it should be reserved for true abuses.
When an AI application falls into what the EU classifies as “unacceptable risk”, namely mass social scoring, psychological manipulation, malicious deepfakes, or subversive disinformation campaigns, the penalties should remain swift and severe. Operating deliberately against human agency and societal stability is unacceptable, and those bad actors deserve heavy fines.
For everything else, from low-risk applications to high-risk industrial models, we need a smarter balance. EU should adopt the US approach of targeted, gated evaluations for high-power general-purpose models, while using positive economic incentives to nurture the rest of the ecosystem.
By pivoting from a fear-based regime of penalties to a value-based model of rewards, Europe could do something remarkable: foster an environment where doing the right thing for citizens is also the most profitable thing for business.
That is how you build an AI strategy that actually lasts.
Yannick HUCHARD