The artificial intelligence sector is witnessing a significant shift in its regulatory landscape as Anthropic, a leading developer of frontier AI models, moves beyond its initial support for transparency laws to advocate for more stringent state-level oversight. Having previously backed the first wave of safety legislation in California and New York—measures that were met with stiff resistance from other Silicon Valley stakeholders—Anthropic now contends that these frameworks are becoming insufficient. The company is currently spearheading a campaign to encourage states to adopt more rigorous mandates, including third-party audits and enforcement mechanisms, as the rapid evolution of AI capabilities outpaces existing policy.
Cesar Fernandez, Anthropic’s head of U.S. state and local government relations, recently emphasized that while the transparency-focused bills of 2025 were a critical milestone, the velocity of technological advancement necessitates a more robust policy response. According to Fernandez, self-reporting and basic disclosure are no longer adequate for the most powerful systems, which possess the potential to impact critical infrastructure and national security. This stance marks a departure from the traditional tech industry ethos of deregulation, positioning Anthropic as a unique proponent of government intervention in a market where it holds a valuation approaching $1 trillion.
The Evolution of State-Level AI Governance
The current push for regulation follows a period of intense legislative activity at the state level, largely driven by a perceived lack of movement in the U.S. Congress. In 2024 and early 2025, California and New York emerged as the primary battlegrounds for AI safety. California’s Senate Bill 53 and New York’s RAISE Act established foundational requirements for developers of "frontier models"—systems that push the boundaries of current computational capabilities. These laws focused primarily on transparency, requiring companies to disclose safety testing protocols and potential risks.
Anthropic’s support for these measures was notable, as competitors like OpenAI and various venture capital firms argued that such requirements would stifle innovation and cede American technological leadership to foreign adversaries. However, Anthropic’s leadership has long maintained that its commercial success is inextricably linked to its founding mission: ensuring a safe transition to transformative AI. This mission-driven approach has led the company to endorse increasingly prescriptive rules as it expands its operations.
In recent months, the focus has shifted from mere transparency to active verification and enforcement. Anthropic has thrown its weight behind an Illinois measure that mandates third-party auditors to evaluate the safety processes of AI labs. More recently, the company endorsed a Massachusetts policy that goes a step further, empowering the state’s attorney general to seek injunctive relief against companies that fail to comply with safety standards. This would effectively grant state officials the power to halt the deployment of specific AI models if they are deemed to pose a significant risk to public safety.
Chronology of Legislative Milestones and Anthropic’s Involvement
To understand the current regulatory climate, it is essential to trace the development of these policies over the last 24 months:
- Early 2024: California introduces initial drafts of safety legislation. Anthropic emerges as a rare industry supporter, while trade groups like NetChoice and the Chamber of Progress launch opposition campaigns.
- September 2025: Governor Gavin Newsom signs SB 53 into law, establishing the first major state-level safety framework for large-scale AI models.
- December 2025: New York Governor Kathy Hochul signs the RAISE Act, mirroring many of California’s requirements but adding specific provisions for models used in critical infrastructure.
- Early 2026: Anthropic hires Cesar Fernandez, formerly of FanDuel and Uber, to lead its state-level lobbying efforts.
- June 2026: Anthropic formally endorses the Massachusetts AI Regulation Bill, marking a pivot toward support for injunctive enforcement powers.
This timeline illustrates a strategic escalation in Anthropic’s policy goals, moving from defensive transparency to offensive safety mandates.
Data and Thresholds: Defining the Scope of Regulation
A central point of contention in the debate over AI safety laws is the definition of which companies should be subject to oversight. Anthropic has consistently advocated for laws that target "large AI model developers." While the specific metrics vary by jurisdiction, the frameworks supported by Anthropic generally apply to companies meeting two primary criteria:
- Development Expenditure: Companies that have invested hundreds of millions of dollars in compute resources and research for a single model.
- Revenue Thresholds: Entities with annual revenues exceeding $500 million.
By focusing on these high-resource tiers, Anthropic argues that the regulations protect the broader startup ecosystem from "red tape" while ensuring that the few companies capable of building potentially catastrophic systems are held accountable. However, critics point out that as capital continues to flood the AI sector, the number of companies meeting these thresholds is likely to grow. Startups such as Safe Superintelligence (SSI), Thinking Machines Lab, and the French firm Mistral have already raised billions of dollars, positioning them as potential subjects of these laws in the near future.
The Debate Over Regulatory Capture
Anthropic’s pro-regulation stance has drawn sharp criticism from prominent figures in Silicon Valley who view the company’s efforts as a form of "regulatory capture." David Sacks, a technology investor and adviser to the Trump administration, has been a vocal critic, alleging that Anthropic is utilizing fear-mongering to build a "moat" around its business. The argument suggests that by advocating for complex and expensive compliance requirements, Anthropic is making it prohibitively difficult for smaller competitors to challenge its market position.
Fernandez has countered these accusations by noting the high thresholds for applicability in the bills Anthropic supports. He maintains that the goal is not to hinder competition but to "inspire a race to the top" where safety is a competitive advantage. Despite these reassurances, the tension between safety-oriented incumbents and growth-oriented challengers remains a defining feature of the AI policy debate.
Further complicating the narrative is Anthropic’s recent interaction with the federal government. Last month, the Trump administration issued a directive requiring Anthropic to suspend access to its "Mythos" and "Fable 5" models for foreign nationals, citing national security concerns. Anthropic expressed dissatisfaction with the move, arguing that such interventions should only occur through a fair and transparent evaluation process. This incident highlights a paradox in Anthropic’s position: while the company advocates for state-level enforcement powers, it remains wary of federal actions that lack what it deems to be sufficient procedural safeguards.
National Security and the Global AI Race
The push for domestic regulation is also influenced by geopolitical competition. Anthropic recently alerted the U.S. government to a "distillation attack" allegedly conducted by the Chinese tech giant Alibaba. In this scenario, Alibaba’s researchers were accused of using systematic prompting to extract internal logic and data from Anthropic’s models to train their own systems.
This event has been used by some AI researchers to argue that Anthropic’s regulatory push has a secondary goal: persuading the U.S. government to restrict or ban open-weight models from foreign adversaries. If successful, such a policy could drive thousands of American businesses—currently utilizing open-source or foreign models—toward proprietary American offerings like those provided by Anthropic and OpenAI. Anthropic CEO Dario Amodei has previously testified before Congress regarding the inherent risks of open-source AI, suggesting that without centralized control, advanced models could be misused by bad actors to develop biological weapons or launch sophisticated cyberattacks.
Broader Societal Impacts and Missing Legislation
While Anthropic has been highly proactive regarding "existential" or "catastrophic" risks, some observers note a discrepancy between the company’s legislative priorities and the concerns of the general public. Recent polling by the Pew Research Center indicates that American voters are primarily concerned with AI-driven job displacement, the environmental impact of massive data centers, and the psychological effects of AI on children.
To date, the coordinated legislative push seen in the safety arena has not been replicated for these socio-economic issues. While Anthropic has published theoretical proposals for addressing job loss and has pledged that taxpayers will not bear the cost of increased electricity demand from data centers, these commitments have not yet translated into the same level of rigorous state-level lobbying. Fernandez has stated that the company is "eager to engage" on these broader topics, but for now, the focus remains squarely on the technical safety of frontier models.
Implications for the Future of AI Policy
As Anthropic continues to shape the conversation, the implications for the technology sector are profound. If more states follow the lead of Massachusetts and Illinois, the U.S. could see a patchwork of stringent AI laws that vary significantly by geography. This "laboratory of the states" approach may eventually force a federal consensus, as tech companies seek a single national standard to simplify compliance.
For the industry, Anthropic’s strategy represents a gamble that safety will become the primary metric by which AI companies are judged. For regulators, it provides a blueprint for how to oversee a technology that is evolving faster than the law can typically keep pace. Whether this leads to a safer digital future or a more consolidated and less competitive market remains one of the most critical questions facing the tech industry in 2026.
About the Author
