# Policy on the AI Exponential "Policy on the AI Exponential" is a June 2026 essay by [[Dario Amodei]], CEO of [[Anthropic]], arguing that AI capability is advancing along an exponential curve while political institutions move at their traditional pace — a mismatch he compares to the hobbits trying to rouse the ponderous Treebeard in *The Lord of the Rings*. The essay marks a significant shift in Anthropic's public policy position: from transparency-focused advocacy (voluntary disclosure, SB 53 in California) to endorsing mandatory third-party testing, government power to block model releases, and a coordinated geopolitical coalition strategy. ## Context and timing The essay was published alongside [[Anthropic]]'s release of a legislative proposal on [[Frontier Model]] testing and a policy framework for job displacement, for which Anthropic committed "substantial financial backing." It arrived days after a Trump administration executive order ("Promoting Advanced Artificial Intelligence Innovation and Security") that moved incrementally toward greater government involvement in AI oversight — a development Amodei acknowledges as positive but insufficient. The immediate catalyst Amodei cites is Claude Mythos Preview and the discovery that frontier models pose real cybersecurity risks — what he describes as proof "beyond doubt that AI models are now tools of global and national strategic consequence." He frames this as the moment when the case for going beyond transparency to binding regulation became undeniable. ## Five policy domains The essay is structured around five policy areas that Amodei argues AI will fundamentally reshape: ### 1. Regulation and public safety The core proposal models AI regulation on aviation (the FAA model). [[Frontier Model]] systems above a compute threshold would undergo mandatory third-party testing in four specific areas: cybersecurity, biological weapons, loss of control of AI systems, and automated R&D that could accelerate the other three risks. The government would have the power to block or reverse deployment if a model presents "unacceptable risks." Testing could be performed by a government agency or by private organizations authorized and inspected by the government — a "regulatory markets" approach. Amodei explicitly distinguishes this from the transparency-only position Anthropic had previously held, explaining that the earlier stance was appropriate when risks were theoretical but is no longer sufficient now that threats are demonstrable. ### 2. Macroeconomics and tax policy Amodei argues that AI may "scramble" the traditional assumption that economic growth is fragile and must be protected from redistributive policy. If AI produces "extremely rapid and robust economic growth" while simultaneously displacing labour at unprecedented scale and speed, the policy challenge shifts from incentivizing growth to ensuring its benefits are shared. He proposes measurement and tracking of AI-driven displacement, pro-employment incentives (wage insurance, retention tax incentives, workforce training grants), and long-term macroeconomic support including universal basic income funded through taxes on AI companies or capital gains taxes. He explicitly acknowledges that enduring displacement may be "an intrinsic property of the technology" rather than a temporary transition. ### 3. Accelerating AI's positive impact Amodei argues that downstream regulatory systems (FDA, EMA) were designed for a slower pace of innovation and risk becoming bottlenecks as AI accelerates drug development, materials science, and energy. He proposes developing standards now for accepting AI simulation in pharmacodynamics, toxicology prediction, dose selection, and synthetic control arms in clinical trials — so the regulatory system is prepared to absorb a deluge of AI-accelerated candidates rather than jamming. ### 4. The state and civil liberties The essay's most striking section warns that AI could be "the ultimate tool of autocracy" and that existing constitutional protections are inadequate. Specific proposals include: accountability rules for fully autonomous weapons systems (requiring them to respond to constitutional authority rather than blindly following orders), a ban on domestic use of fully autonomous weapons, closing the bulk data collection/data broker loophole, and guaranteeing citizens access to AI "at least as capable" as whatever the government uses against them in adverse actions. Amodei also warns that AI companies themselves could "capture the state or adopt quasi-state characteristics" and argues for separation-of-power mechanisms within AI companies — citing [[Anthropic]]'s Long-Term Benefit Trust as one model. ### 5. Securing leadership by democracies The geopolitical section argues that AI is not merely an instrument of trade policy but "something that resets the whole game board" — comparable to nuclear weapons but potentially more consequential. He proposes a democratic coalition that coordinates AI supply chain management (sharing chips within the coalition, denying them to adversaries), harmonized regulation, shared benefits, mutual defense, rejection of AI-powered repression, and macroeconomic cooperation. The vision is explicitly expansionist: the coalition should "iteratively" draw in the rest of the world by making membership attractive and exclusion costly. ## The policy shift For three years, [[Anthropic]]'s public policy position was transparency — voluntary disclosure, supporting legislation like California's SB 53, and arguing that binding regulation was premature because the specific shape of risks was unclear. "Policy on the AI Exponential" abandons that position explicitly: "now the risks are clearly here. It is time to go beyond transparency to more serious and binding regulation of AI." This shift is significant because Anthropic had positioned itself as the "responsible" AI company — more cautious than [[OpenAI]] or Google, willing to accept slower deployment for safety. The essay represents the moment when the safety-first company concluded that voluntary safety was insufficient and called for regulatory authority with enforcement power. ## KB significance The essay matters for the vault on several levels. First, it represents the most comprehensive policy framework from an AI company CEO, covering safety, economics, civil liberties, and geopolitics in a single document. Second, the autonomous weapons section directly engages with the [[Autonomous Weapons Debate]] and [[Dual-Use AI]] themes the vault tracks — and takes a notably more cautious position than the defense-tech ecosystem's "move fast and deploy" ethos. Third, the democratic coalition proposal is essentially a framework for a new Cold War alignment structured around AI capability — connecting to the vault's coverage of the [[Chinese Space Program Hub]], [[New Cold War]], and great-power competition. The essay has been criticized from multiple directions. Defense-tech accelerationists view the regulatory proposals as competitive self-handicapping. AI safety advocates note that Anthropic is proposing to regulate a market it dominates — the "regulatory capture" concern. Civil libertarians observe that the autonomous weapons ban applies only domestically, implicitly accepting their use abroad. And the labour proposals, while unusually candid about displacement risk for a tech CEO, offer no mechanism for worker agency or collective bargaining. The Treebeard metaphor is revealing: Amodei casts democratic institutions as inherently slow and AI companies as the urgent voices of reason — a framing that assumes the hobbits (tech companies) correctly understand the threat and the trees (governments) need to be activated on the hobbits' terms. Whether this reflects genuine public interest or a sophisticated bid to shape the regulatory environment in [[Anthropic]]'s favour is the essay's central unresolved question. ## See Also - [[Dario Amodei]] — the author - [[Anthropic]] — the company - [[Frontier Model]] — the regulatory target - [[Autonomous Weapons Debate]] — the civil liberties dimension - [[Dual-Use AI]] — the military application question - [[Situational Awareness]] — Leopold Aschenbrenner's earlier essay with overlapping themes - [[The Singularity]] — the exponential trajectory Amodei describes