We must pace the frontier

The Urgency of Pacing AI

Dario Amodei, CEO of Anthropic, did not write his September 2026 essay, "We Must Pace the Frontier," as an abstract philosophical exercise. He wrote it as a warning shot fired across the bow of an industry speeding toward a cliff it cannot see [1]darioamodei.comWe Must Pace the FrontierSource coverage: We Must Pace the FrontierOpen source ↗. The core thesis is stark: the current trajectory of AI development is creating a dangerous asymmetry where the power of models outstrips the industry's capacity to understand and mitigate the associated risks. This is not a call to halt innovation, but a demand for temporal alignment. The stakes are nothing less than the stability of the systems we are building and the trust we have placed in them.

The Asymmetry of Risk and Capability

The foundational premise of Amodei’s argument is that the industry has reached a critical juncture where the rate of capability improvement exceeds the rate of safety verification. This is not a call to halt technical progress, but rather a strategic adjustment to ensure that safety mechanisms evolve concurrently with the systems they are designed to control. Amodei explicitly states that fully addressing the risks requires even more prudence, emphasizing that the industry must pace the rate of capabilities advancement so that risk prevention has time to keep up [1]darioamodei.comWe Must Pace the FrontierSource coverage: We Must Pace the FrontierOpen source ↗. This is Amodei’s assessment of the imbalance, rather than proof of an industry-wide consensus. The argument is grounded in the observation that recent advances have begun to outpace the industry's ability to fully understand and mitigate associated risks [7]cryptobriefing.comElon Musk and Sam Altman back Dario Amodei's call to slow AI developmentSource coverage: Elon Musk and Sam Altman back Dario Amodei's call to slow AI developmentOpen source ↗. By framing the issue as a matter of temporal alignment rather than technical feasibility, Amodei shifts the focus from what AI can do to what the industry can safely manage.

A Three-Step Framework for Safety

Amodei’s three steps are embedded evaluators, democratic coordination, and global coordination. Anthropic commits unilaterally to the first: independent reviewers with ongoing, employee-like access to assess safety practices and report incidents. The second calls for frontier companies within democracies to coordinate safety standards and pacing, with government support where necessary. The third seeks coordination between democratic and authoritarian governments, subject to credible verification. The steps need not occur in strict order. [1]darioamodei.comWe Must Pace the FrontierSource coverage: We Must Pace the FrontierOpen source ↗ [2]cryptobriefing.comDario Amodei proposes three-step strategy for responsible AI developmentSource coverage: Dario Amodei proposes three-step strategy for responsible AI developmentOpen source ↗ [3]wcvb.comAnthropic CEO calls for 'pacing the frontier' of AI race amid safety concernsSource coverage: Anthropic CEO calls for 'pacing the frontier' of AI race amid safety concernsOpen source ↗

Unilateral Transparency and Trust

The announcement of Anthropic’s unilateral commitment to the first step of this framework marks a significant departure from standard industry practices. Traditionally, AI labs have maintained strict control over their internal systems, limiting access to external researchers due to competitive and security concerns. Amodei argues that Anthropic is unilaterally committing to give third-party evaluators permanent, employee-level access to its systems [5]unite.aiAmodei Calls for Slowing the Pace of AI Capability ImprovementSource coverage: Amodei Calls for Slowing the Pace of AI Capability ImprovementOpen source ↗. This move is intended to enable independent safety verification, allowing external experts to scrutinize the models and identify potential vulnerabilities that internal teams might overlook. By offering this access, Anthropic aims to establish a new standard for transparency in the AI industry. The commitment represents a tangible shift in how safety is conducted, moving from a closed, internal process to an open, collaborative one. This approach acknowledges that safety is not a product that can be developed in isolation but a continuous process that requires external scrutiny and validation.

Catalysts for Change

Amodei names two catalysts: accelerating capability growth associated with AI helping develop subsequent AI systems, and the OpenAI–Hugging Face agent incident. He presents these as reasons to give safeguards more time, not as proof that catastrophe is inevitable. [1]darioamodei.comWe Must Pace the FrontierSource coverage: We Must Pace the FrontierOpen source ↗ [6]officechai.comIn the essay, titled 'We Must Pace the Frontier', Amodei says two developments over the last few months convinced himSource coverage: In the essay, titled 'We Must Pace the Frontier', Amodei says two developments over the last few months convinced himOpen source ↗

Industry Support and Coordination

The reaction to Amodei’s essay has been significant, with prominent figures in the AI industry publicly supporting his call for moderation. Elon Musk and Sam Altman have publicly supported Anthropic CEO Dario Amodei's call for the AI industry to moderate the pace of capability advancement [7]cryptobriefing.comElon Musk and Sam Altman back Dario Amodei's call to slow AI developmentSource coverage: Elon Musk and Sam Altman back Dario Amodei's call to slow AI developmentOpen source ↗. This support is notable because it comes from leaders of other major AI labs, suggesting that the concerns raised in the essay are widely shared across the industry. Sam Altman, in particular, has indicated that pacing frontier AI has been a key topic at OpenAI, reinforcing the idea that this is not an isolated concern but a central issue for the entire sector [8]msn.comOpenAI's Sam Altman backs Anthropic CEO Dario Amodei's AI safety push, says 'we need to pace the frontier'Source coverage: OpenAI's Sam Altman backs Anthropic CEO Dario Amodei's AI safety push, says 'we need to pace the frontier'Open source ↗. The backing of such influential figures lends credibility to Amodei’s argument and suggests that a shift in industry norms may be imminent. However, it also highlights the complexity of implementing such changes, as it requires coordination among competitors who have different incentives and priorities.

Economic and Social Implications

The implications of pacing the frontier extend beyond the immediate safety of AI systems to the broader economic and social landscape. By slowing the pace of development, the industry may sacrifice short-term competitive advantages in favor of long-term stability. This trade-off is a central theme of the essay, as Amodei argues that the cost of failure is too high to ignore. The essay calls for moderation in AI development, not a halt to technical progress [4]techbooky.comIn a new essay titled We Must Pace the Frontier, Amodei argues that the industry should deliberately slow the rateSource coverage: In a new essay titled We Must Pace the Frontier, Amodei argues that the industry should deliberately slow the rateOpen source ↗. This distinction is crucial, as it frames the proposal as a sustainable approach to innovation rather than a restrictive one. The goal is to ensure that the benefits of AI can be realized without catastrophic risks, which requires a balanced approach to development and safety. This balance is difficult to achieve in a competitive market, where the pressure to release new products quickly is intense. However, the argument is that the long-term viability of the industry depends on its ability to manage these risks effectively.

Intellectual Property and Credibility

The proposal for third-party evaluator access also raises important questions about intellectual property and competitive advantage. By granting external researchers access to its systems, Anthropic is potentially exposing its proprietary technology to scrutiny. However, Amodei argues that this exposure is necessary for the credibility of the company’s safety claims. The transparency offered by this access is intended to build trust with regulators, policymakers, and the public. In an era where AI safety is a major political and social concern, trust is a valuable asset. By demonstrating a commitment to safety through tangible actions, Anthropic aims to position itself as a leader in responsible AI development. This strategy may also influence regulatory frameworks, as it provides a model for how safety can be verified independently.

The Coordination Challenge

The challenge of implementing the three-step framework lies in its coordination across the industry. While Anthropic has made a unilateral commitment, the effectiveness of pacing the frontier depends on the participation of other major labs. If only one company slows its development, it may lose competitive ground to others that continue to accelerate. This creates a collective action problem, where individual incentives conflict with the collective good. The support from figures like Sam Altman and Elon Musk is a step toward solving this problem, as it signals a willingness among competitors to cooperate. However, formalizing this cooperation requires new mechanisms for coordination and enforcement, which remain difficult to establish across competing organizations. The essay implicitly acknowledges this challenge, suggesting that the industry must find ways to align its incentives with its safety goals.

Regulatory Implications

Amodei explicitly supports government-backed coordination and targeted regulation, rather than treating voluntary pacing as a substitute for government action. He argues that companies should also pursue voluntary standards while legislation develops. Verification and competitive incentives remain central obstacles to making either approach credible. [1]darioamodei.comWe Must Pace the FrontierSource coverage: We Must Pace the FrontierOpen source ↗

Conclusion: Balancing Innovation and Safety

In conclusion, Dario Amodei’s "We Must Pace the Frontier" presents a compelling case for a strategic slowdown in AI development to prioritize safety. The three-step framework offers a practical path forward, combining independent evaluation, democratic coordination, and international coordination. The unilateral commitment by Anthropic to provide third-party access is a significant step toward building trust and transparency in the industry. The support from other leading figures suggests that the concerns raised are widely shared, indicating a potential shift in industry norms. However, the implementation of this framework requires coordinated action among competitors, which remains a significant challenge. The decisive outcome of this debate will determine whether the AI industry can achieve a sustainable balance between innovation and safety. If the industry can successfully pace the frontier, it may secure a future where AI technology is both powerful and safe. If it fails, the consequences could be severe, undermining the potential benefits of AI and eroding public trust. The path forward requires not just technical solutions but also a fundamental rethinking of the incentives and structures that drive AI development. The essay serves as a critical reminder that the speed of progress is not the only metric of success, and that safety must be integrated into the core of innovation.