The Carolina Principles: America's Soft-Power Gambit for AI's Global Rulebook

The protocol remembers what the regulators forget. But this September, the regulators are writing the protocol. G20 Innovation Ministers gather in North Carolina on September 1-2. On the table: the "Carolina Principles" — a non-binding framework that could quietly redraw the global AI governance map. Not through treaty obligations. Not through enforcement mechanisms. Through narrative capture. The lineup reads like a who's who of American AI power: Elon Musk, Sam Altman, Jensen Huang, and David Sacks — the White House's AI & Crypto Czar. This is not diplomacy. This is industrial policy wearing a diplomatic costume. And the crypto world should be watching closely, because the playbook is identical to what we've seen in digital asset regulation: soft law first, market capture second, enforcement third. The Carolina Principles rest on three pillars: avoid creating new regulatory bodies, rely on existing industry regulators, and allow government-enterprise joint testing of new technologies. Each pillar is a direct repudiation of the EU's AI Act — the risk-tiered, compliance-heavy framework that took effect in August 2024. This is the Trump administration's domestic policy, internationalized. In March 2025, the White House explicitly declined to establish a federal AI regulatory agency. Now that same philosophy is being exported to the G20, where consensus-based decision-making means a few countries can't easily veto the agenda. The strategic logic is clear. The EU chose hard law — binding regulations with teeth, including fines up to 7% of global revenue. The US chose soft law — non-binding principles that are easier to adopt, easier to adapt, and easier to ignore when inconvenient. Soft law wins adoption races. Hard law wins compliance battles. The US is betting that adoption is the more valuable currency. My own experience in Vienna's regulatory trenches taught me this lesson. During the MiCA negotiations, I watched how non-binding frameworks shaped the final text more than any formal position paper. The Austrian Data Privacy Regulatory Lobby campaign succeeded not through confrontation, but through narrative alignment — getting the right people in the room to agree on principles before the details were drafted. The Carolina Principles are that same playbook, scaled to the G20. Here's what the framework actually does, stripped of diplomatic language. First, it lowers compliance costs for AI enterprises. The EU's AI Act imposes compliance burdens estimated at 1-3% of revenue for high-risk systems. The Carolina Principles eliminate that drag. For unlisted AI giants like OpenAI, Anthropic, and xAI — whose valuations depend on optimistic regulatory assumptions — this is a direct tailwind. The financial engineering here is elegant: reduce regulatory friction, increase cash flow expectations, watch the multiples expand. Second, it creates a "coalition of the willing" against EU regulatory influence. Japan, South Korea, India, Australia — all AI-friendly jurisdictions with reasons to prefer light-touch governance. The US is building an institutional counterweight to Brussels, and the G20 is the venue where that counterweight gains legitimacy. This is the same dynamic we saw in crypto: when the US pushed for lighter digital asset rules, it found allies in Singapore, the UAE, and Switzerland. The EU's maximalist approach alienated the very jurisdictions that wanted to build. Third, it signals to capital markets that American AI has political backing. Musk, Altman, and Huang aren't attending as private citizens. They're attending as the industrial base of American AI — applications, models, and compute respectively. Their presence tells global investors: the US government and its AI sector are aligned. NVIDIA's valuation, OpenAI's fundraising, xAI's expansion — all benefit from this perceived alignment. The market reads attendance as endorsement, and endorsement as reduced political risk. But here's the part the official narrative misses. The "government-enterprise joint testing" provision is a data acquisition channel. When governments co-test AI systems with enterprises, they gain access to technical details, training data, and deployment architectures. Under the banner of national security, this is justifiable. But it also creates a mechanism for industrial intelligence gathering that has nothing to do with safety. The same logic applies to crypto: when regulators demand access to exchange data "for consumer protection," they're also building surveillance infrastructure. And the naming matters. "Carolina Principles" — named after the host location, echoing the Bretton Woods system. That's not accidental. It's a claim to historical significance, an attempt to position this framework as the foundational governance document of the AI era. Bretton Woods gave us the IMF and the World Bank. The Carolina Principles, if successful, would give us a global AI governance architecture centered on American interests. Now the uncomfortable part. The light-touch approach has a fatal assumption embedded in it: that American AI companies can self-regulate effectively. The evidence says otherwise. OpenAI's hallucinations, Google's bias incidents, the data leaks across major labs — market discipline has not prevented these failures. The "race to the bottom" is not a hypothetical. It's the natural equilibrium of uncoordinated regulatory competition. If every major AI jurisdiction adopts light-touch governance, the incentive structure pushes toward cutting safety corners to ship faster. I've seen this pattern in DeFi: when protocols compete for TVL without safety standards, the ones that cut corners win — until they collapse. The accountability vacuum is real. Non-binding principles mean no enforcement mechanism. When AI causes cross-border harm — and it will — there's no clear liability framework. The EU's AI Act, for all its bureaucratic excess, at least provides a penalty structure. The Carolina Principles provide... principles. That's not governance. That's a wish list. And the global South gets left out entirely. Developing countries lack the technical capacity for AI safety assessment. If they adopt light-touch frameworks, they become the weak link in global AI governance — the jurisdictions where unvalidated systems get deployed first. The framework's silence on this issue is telling. It's not an oversight. It's a design choice. The December G20 Leaders Summit is the real test. If the Carolina Principles make it into the joint declaration, they gain the highest level of international political endorsement. If not, the framework remains a ministerial footnote — a nice document that nobody is obligated to follow. Speed without direction is just volatility. The US is moving fast to define AI's global rulebook. The question is whether the direction serves innovation, safety, or simply American market dominance. Regulation is the friction that forces efficiency — but only when the friction is designed correctly. Watch the December text. That's where the future gets written.