Anthropic, not a neutral AI lab
A clear thesis helps. Anthropic is not a neutral research shop, it is a political project that advances the Biden‑era AI program after voters rejected it in 2024. The mechanism is simple. Staff the company with the authors of the prior regime, fund it with Democrat mega-donors, and route policy through blue state capitols when the White House changes hands. That is David Sacks’ claim. To steelman it we assemble the strongest case, then test objections. The conclusion holds. On personnel, legislation, lobbying, and money, Anthropic functions as a de facto resistance outfit, one that aims to codify progressive norms inside AI systems and to bind the rest of the country to California’s model.
Begin with the contrast. Trump’s second‑term program for AI is deregulatory, pluralist, and growth‑first. It rejects the Biden framework that centered civil‑rights style mandates, bias audits, and broad preclearance concepts. Trump scrapped that framework in January 2025, removed the diffusion export rule that yoked chips, weights, and cloud capacity to a tiered control scheme, and replaced it with an innovation‑forward posture that prizes speed, industrial capacity, and ideological neutrality inside deployed models. The Trump view, in brief, is that state, market, and scientific competition will discipline risk while American firms race. The rival view, embraced by Anthropic, is that safety, equity, and precaution must be preemptively codified and enforced through governance. These are different philosophies of both technology and politics.
California as backdoor Congress
When a federal program falls, its architects seek another venue. California is the world’s most powerful subnational regulator of technology. If Sacramento imposes a rule on large models, any firm with products or staff in California must either comply or litigate at high cost. So a California rule becomes a national rule in practice. That is the backdoor. In 2025, Anthropic publicly endorsed California’s SB 53, the Transparency in Frontier AI Act. SB 53 requires developers of frontier‑scale models to publish safety frameworks, risk assessments, and transparency reports. Its language foregrounds fairness, accountability, and responsible AI. Its sponsors described it as a model for the nation and a bridge in the absence of federal action. One may welcome those goals, but the structure is plain. If the federal center removes a mandate, the California periphery reinstates it for everyone who touches the state.
The pattern is wider than a single bill. California’s legislature has repeatedly advanced measures that embed algorithmic fairness and due process into automated decision making, including notice, appeal, and anti‑discrimination rules. Some stalled, others moved forward, but the direction is consistent. It grafts civil‑rights enforcement logic onto AI tooling. Anthropics support for SB 53, public and emphatic, placed the company in active alignment with that program. The effect is a blue‑state lever applied to a national industry, a lever that bypasses the federal reset chosen by voters in November.
People are policy
An organization reveals its aims through its hires. In early 2025 Anthropic recruited Elizabeth Kelly, tapped by President Biden to lead the U.S. AI Safety Institute. It brought in Tarun Chhabra from the National Security Council, a principal advocate for stringent export controls, and retained Ben Buchanan, a Biden White House advisor on AI, as a senior counselor. These are the architects of the prior regime. Their expertise is real, their public service is not in question, but their policy priors are equally real. If you staff your policy shop with the same people who wrote the last administration’s rules, it is reasonable to expect you will pursue those rules by other means.
The export story shows this cleanly. Biden’s diffusion rule attempted to bind chips, model weights, and cloud compute with a tiered licensing system. Trump’s team rescinded it as complex, innovation‑dampening, and misaligned with a coalition strategy that accelerates allied capacity. Anthropic criticized the rollback, urged faster and stricter controls, and warned that delay would invite adversary stockpiling. The merits of export control are debatable. The point is political alignment. The company closed ranks with the prior administration’s hawkish technonationalism, and against the current administration’s pro‑production turn. When a private firm persistently takes the side of the ousted program against the sitting one, the word opposition is ordinary English, not hyperbole.
Lobbying against Trump’s One Big Beautiful Bill
Personnel and statements matter, but law is where agendas cash out. Trump’s flagship AI legislation, widely dubbed the One Big Beautiful Bill, initially included a 10‑year preemption that barred states from enacting their own AI regimes. The purpose was to prevent a patchwork that would chill investment and balkanize compliance. Anthropic fought the preemption. Its executives urged Congress to strip it, its allies argued in op‑eds for transparency mandates that resembled SB 53, and sympathetic members echoed the state‑rights line. Preemption was removed. Blue states retained full runway to pass AI codes, exactly the runway Anthropic had celebrated in California. If you wish to impose progressive guardrails on AI despite a deregulatory federal turn, this is the move you make. You preserve the venue where your allies have power, then you help them legislate.
The donor network
Funding is not dispositive, but it is suggestive. Anthropic’s cap table and board feature the highest tier of Democratic political money. Reed Hastings, a major donor to Democratic causes, joined the board in 2025. Dustin Moskovitz, among the largest funders of Democrat committees in 2024, is an early investor. Eric Schmidt’s family office participated early. Sam Bankman‑Fried invested heavily in 2021. The company’s leadership, including CEO Dario Amodei, has made public statements hostile to Trump and donations to Democratic candidates. None of this is illegal or unusual in Silicon Valley, but it does falsify the neutrality claim. If a media outlet were staffed by former administration officials and funded by partisan donors, no one would hesitate to call it aligned. The same inference is rational here. Money and message point in the same direction.
What about safety
A common objection is that Anthropic is not partisan, it is safety‑first. Safety, the objection continues, is orthogonal to politics. This sounds attractive. It also evades the basic question. Which safety, and defined by whom. A framework that centers fairness audits, demographic outcomes, and speech policing inside foundation models is a framework with political content. It encodes a theory of justice and of culture. A framework that centers reliability, robustness, and catastrophic risk reduction while maintaining viewpoint neutrality is another theory with different political consequences. Choosing between these is not apolitical. It is a choice about the values the state will impose on general purpose tools. Anthropic’s record reveals its preference. It pushed California to mandate transparency and bias reporting, it stacked its policy team with the authors of the Biden program, it fought federal preemption that would have left innovation to markets and federal standards, and it campaigned for strict export controls even when the White House shifted course. That is not value free inquiry. That is a policy program.
The federalism dodge
Another objection appeals to federalism. Why, the critic asks, should conservatives oppose states that wish to protect their citizens through AI guardrails. Federalism is good, but selective federalism is not. The question is not whether California can regulate within its borders. The question is whether one state should use its market power to dictate the national template after a federal election has settled the basic direction. Conservatives support federal preemption when states attempt to nullify constitutional rights or to erect barriers that distort national markets. The Trump bill attempted to avoid a balkanized AI regime precisely because the cost of compliance scales with model complexity. Leave the rules to fifty capitols, and only the largest incumbents can survive the maze. That is what blue‑state progressives claim to fear about industry concentration, yet it is what their policy demands produce.
The neutrality principle
The Trump administration has insisted that government AI systems must be ideologically neutral. This is not censorship, it is the opposite. A federal model should not encode the metaphysics of gender advocated by one party, nor should it degrade traditional religious speech under the banner of safety. It should provide factual competence without moralizing. By contrast, the Biden‑era program, echoed in California’s language, integrates equity targets, speech norms, and outcome equalization into the technical stack. That approach, whatever its intentions, drifts toward centralized control of permissible outputs. It is no answer to say that private firms can build what they like. When the state subsidizes, purchases, and certifies models that enforce a comprehensive moral vision, private actors are pulled into the same channel by liability and procurement pressure. The neutrality principle resists that pull.
Wider stakes, not just a fight over jargon
The stakes extend beyond regulatory taste. AI is an instrument of national capacity. Chips, models, and cloud buildouts are capital intensive, and their returns compound. A fast, permission‑light environment, one that clears federal obstacles and bars state duplication, will favor open competition and scale. A slow, permissioned environment, one that disperses authority across blue‑state boards and commissions, will favor incumbents that can staff compliance departments and lobby committees. The former path is compatible with a country that wants to out‑innovate rivals while protecting core civil liberties. The latter path is compatible with a country that wants to harmonize speech and social outcomes through technical levers. That is the actual divide, and Anthropic has chosen a side.
Answering the best objections
Is the case overstated because some of Anthropic’s positions, such as tough export controls, coincide with hawkish Republican views. Not necessarily. Bipartisan overlap on a single axis does not negate the broader alignment. The question is whether Anthropic used its influence to reinstall the Biden architecture after the public repudiated it. Its work against preemption, its endorsement of California’s template, and its staffing choices show consistent intent. Is it unfair to describe the company as Resistance when it performs valuable research and publishes safety results. No. Work product does not erase political action. A university can conduct real science and still run a political shop inside its administration. The relevant point is institutional function in the policy arena, not technical competence in a lab.
A constructive path
The point is not to anathematize a single company. It is to clarify the policy boundary. The Trump administration should continue to do four things. First, enforce federal neutrality for models used by government, so that citizens encounter services that answer to law and fact, not to partisan ideology. Second, keep federal rules focused on catastrophic risk, reliability, and security, while declining to hard code contested cultural norms into the technical stack. Third, preempt state AI codes that export one state’s morality to the nation under the banner of safety, and replace them with narrow, technology neutral standards that protect consumers without freezing innovation. Fourth, couple deregulation with production, including rapid permitting for fabs, data centers, and transmission, so that the US wins the scale game in chips and compute.
A candid word to Anthropic
If Anthropic wants to be seen as nonpartisan, it has a straightforward option. Stop using California as a national policy machine, stop importing one party’s policy staff as a turnkey government in exile, and stop painting deregulatory moves as reckless when they are simply different. Make safety proposals that do not smuggle in contested social agendas. Support a single national floor on catastrophic risk and security while letting markets and civil society adjudicate the rest. If the company chooses the opposite course, it should be candid with the public. It is a political actor, not just a lab, and it prefers the Biden program to the Trump program. In a free country, it can make that choice. In a free country, others can recognize it for what it is.
Conclusion
David Sacks’ charge, properly steelmanned, stands. Anthropic is aligned in personnel, lawfare, and funding with the prior administration’s AI regime, and it has used blue state leverage to reinstall that regime after a federal reversal. Trump’s program, by contrast, aims to restore neutrality, accelerate production, and keep the state from writing ideology into the weights of general models. The choice is bright. It is between a world where California’s boards decide what national AI must look like, and a world where the U.S., under a deregulatory federal policy, builds faster than rivals while respecting viewpoint diversity. Voters settled this once already. The administration is right to fight back and keep the victory meaningful.
If you enjoy my work, please share my work and subscribe: https://x.com/amuse.
READ NEXT: Fox Star Loses It On Fmr GOP Rep On Live TV
Defund, NO grants