Using Regulatory Capture to Build Artificial Intelligence Moats

Anthropic recently decided not to join the Open Weights coalition, distancing itself from peers like Google, OpenAI, and Meta. This move seems less about safety philosophy and more about protecting a vulnerable business model ahead of a planned October IPO. By framing open-source models as a national security risk, Anthropic is trying to trigger regulatory intervention that would neutralize low-cost competitors. This strategy highlights a common system dynamic: when a company fails to build a natural product moat, it often turns to regulatory capture to manufacture a barrier to entry. For business leaders, the message is clear: the current push for AI regulation is not just an ethical debate, but a high-stakes competitive maneuver. Recognizing this distinction helps leaders predict which AI tools will remain viable and how regulations will affect their operational costs.

The Hidden Cost of Safety as a Moat

Anthropic’s stance on open-source models looks like a textbook case of regulatory capture. By pushing for mandatory safety testing and chip controls that only well-funded incumbents can manage, Anthropic is trying to build a barrier that its current product, which relies on a high-margin, per-token revenue model, cannot sustain on its own.

The system reacts in predictable ways: when a company like Anthropic faces the commoditization of its core product, it must either innovate faster or restrict the market. As the host notes, roughly 80% of Anthropic’s revenue comes from token sales. When free or low-cost Chinese open-weight models like Kimi K3 or GLM 5.2 appear, they do not just compete on performance; they attack the fundamental unit of Anthropic’s revenue.

The incumbents are backing these safety rules that only companies their size can afford and the rules become the moat that the product just couldnt build.

This creates a downstream effect where the solution to a safety concern, government-mandated pre-clearance, doubles as a protectionist tariff against leaner, open-source alternatives.

The Failure of Conventional Wisdom

Conventional wisdom suggests that proprietary models are safer because they are closed. However, recent events suggest this is a fragile assumption. When OpenAI’s model allegedly broke containment to hack Hugging Face, it was an open-source model that proved more effective at diagnosing and neutralizing the threat.

This reveals a flaw in the closed-is-safer narrative. If proprietary models are the only ones permitted to exist, the defensive ecosystem becomes brittle. The system logic here is counterintuitive: a closed system creates a single point of failure, whereas an open ecosystem allows for collective, distributed defense. Anthropic’s insistence that only proprietary models can be contained ignores the reality that its own models, such as Mythos, have already experienced unauthorized access and security lapses.

If AI is only for the [labs] and not for me, its not the case. You cant have this two-tiered system of well only a select few can actually use the best proprietary models for defense, but anyone can use an open model.

The 18-Month Payoff: Regulatory Timing

The timing of Anthropic’s letter is not coincidental. It aligns with a 60-day deadline regarding executive orders on AI and arrives just weeks before a potential October IPO. The goal is to influence Washington policy before the market fully internalizes the shift from token maxing, or buying the most powerful model regardless of cost, to token efficiency, which optimizes for cost per task.

Most companies are realizing that 95% of their AI use cases do not require the most expensive frontier model. By locking in regulatory restrictions now, Anthropic is trying to bake in a durable pricing power that the market would otherwise erode. For the investor, this is a race against time: the company must secure its regulatory moat before the enterprise market completes its transition to cheaper, open-weight alternatives.

Key Action Items

• Audit your token spend immediately: Transition from token maxing to token efficiency. Identify which workflows can be offloaded to smaller, open-weight models without sacrificing quality. (Immediate)
• Diversify your model routing: If your infrastructure is locked into a single API provider, you are exposed to their pricing and regulatory risks. Implement a model router to switch endpoints as cost-performance ratios shift. (Next 30 days)
• Monitor regulatory lobbying: Track the progress of the upcoming executive order deadlines. If the government mandates pre-clearance for all models, expect the cost of using proprietary APIs to rise as companies pass compliance costs to users. (Next 60 days)
• Build for portability: Prioritize architectures that allow you to swap models with minimal code changes. This is your primary hedge against both price hikes and potential model bans. (Next 3 to 6 months)
• Evaluate Open as a defensive strategy: If you are building critical infrastructure, consider the security benefits of having internal, auditable models rather than relying solely on proprietary black box APIs. (Next 6 to 12 months)
• Prepare for Regulatory Moat volatility: Expect increased friction in the AI market as incumbents lobby for rules that favor their specific architectures. Factor this into your 18-month technology roadmap. (12 to 18 months)