The Current Reality of American AI Policy: From ‘Pause AI’ to ‘Build’

The discourse around AI in the United States has undergone a dramatic cultural and political shift, moving from a position of fear and calls to “pause AI” to a proactive strategy focused on winning the global AI race. This evolution was driven by a realization that the initial narrative, captured by “doomerism,” was creating a dangerous chilling effect on innovation, especially as adversaries like China made rapid progress.

The “Pause AI” Era and Flawed Narratives

Just a year or two ago, the dominant conversation was shaped by fear. The Biden administration’s initial executive order was seen as limiting innovation, and this sentiment was amplified by technologists and prominent figures signing petitions about existential risk. This stood in stark contrast to previous tech waves like the internet, where even in the face of concrete dangers like the Morris Worm, the national doctrine was to accelerate investment and development.

This period of caution culminated in proposed legislation like California’s SB 1047, which was seen as a particularly egregious example of misguided policy. The bill proposed holding developers of open-source models liable for downstream misuse, a move that would have been absurd and created a massive “chilling effect” on research and development. The core issue was a departure from 40 years of established tech policy without sufficient evidence to justify such a change. The prevailing view was that “extraordinary claims require extraordinary evidence,” and the burden of proof should be on those arguing for drastic, innovation-stifling measures.

Deconstructing the Arguments Against Open Source

A central battleground was the debate over open source AI. The arguments against it were often based on flawed analogies and theoretical risks:

• Flawed Analogies: Opponents equated open-sourcing AI models to publishing plans for nuclear weapons or F-16 fighter jets. This fundamentally misunderstood the nature of the technology. An F-16 is a weapon, whereas AI is a dual-use, general-purpose technology, more akin to a jet engine. The focus was on weaponized applications rather than the foundational tech.
• Theoretical Risks: Arguments often centered on hypothetical dangers, such as AI creating bioweapons, without empirical evidence that these models introduced significant marginal risk beyond what was already possible with existing tools like Google Search.
• The US Lead Fallacy: A key atmospheric argument was that the US was years ahead of China, so locking down technology would maintain this lead. This was a factual inaccuracy. The rapid emergence of powerful models from China, such as DeepSeek, shattered this illusion and served as a catalyst, making it clear to policymakers that the US was in a competitive race. As one speaker noted, “it felt like we were being gaslit constantly because both the content and the atmospherics were just wrong.”

Open Source as a Business Strategy

The conversation has now matured to recognize open source not just as a philosophical or political stance, but as a potent business strategy, following patterns seen in previous infrastructure waves (e.g., Linux and Red Hat).

AI presents a unique, more sustainable version of the “open core” model:

1. Distribution and Ecosystem: Releasing smaller models with open weights is a powerful way to gain distribution, build a brand, and foster a developer ecosystem.
2. IP Protection: Unlike traditional open-source software, releasing model weights does not give away the entire “factory.” The true intellectual property lies in the massive datasets, data pipelines, and post-training techniques, which remain proprietary.
3. Monetization: Companies can offer their largest, state-of-the-art models via paid APIs while using the open-source models to build a funnel. This allows them to capture enterprise and government customers, especially in regulated industries or for sovereign AI use cases where on-premise deployment and control are critical.

Ultimately, the long-term ecosystem win—having the world run on American models, chips, and techniques—is orders of magnitude more valuable than any perceived short-term loss of IP.

The AI Action Plan: A Pragmatic Step Forward

The new AI Action Plan represents a marked shift, starting with its inspirational framing: “Today, a new frontier of scientific discovery lies before us.” The plan is praised for its pragmatism and sophistication.

Key Strengths:

• Technologist Involvement: It was co-authored by technologists, bridging the gap between Silicon Valley and DC.
• Acknowledges Open Source: It recognizes open source as a strategic necessity for winning the AI race.
• Focus on Evaluation: It proposes building an “AI evaluations ecosystem,” a sophisticated approach that prioritizes creating a scientific, grounded framework for measuring risk before jumping to regulation.

Critiques and Next Steps:

• Lack of Execution Details: The plan is ambitious but light on the “how.” The next phase requires the hard work of implementation.
• Omission of Academia: There is a notable lack of focus on investing in academia, a traditional engine of American innovation.

Rethinking Risk and the Path Forward

The discourse is moving toward a more nuanced understanding of risk.

• Marginal Risk: Instead of treating all AI risks as novel, the concept of marginal risk asks: does AI introduce new dangers that our existing, well-honed frameworks for managing complex technological systems cannot handle? If not, we should apply what we’ve learned over decades rather than inventing entirely new liabilities.
• The Cost of Inaction: The “what’s the rush?” argument is countered by the immense opportunity cost of slowing down. Every month lost is a delay in potentially solving major challenges in science and medicine. The probability of doom, or “p-doom,” without AI’s benefits may be far greater than with it. The consensus is that while we must explore the implications of AI on jobs, defense, and safety, this exploration should not halt progress. The race is on, and the new posture is to ensure the US leads it.

• Tags:
• AI Safety
• Feature
• Geopolitics
• AI Policy
• Innovation
• Regulation
• Open-Source AI
• SB 1047