Earlier this month, I warned that the coming AI fortunes could turn effective altruism into a permanent source of political power. Anthropic's latest announcement gives that warning a dollar figure.

On October 18, the company proposed committing 5% of its shares to a new Public Benefit Fund, subject to stockholder approval. Combined with its founders' pledges to donate 80% of their own equity and its employee giving program, Anthropic says roughly 20% of the company would be dedicated to the public good.

Effective altruism's ugliest political temptation is the conviction that superior moral arithmetic entitles its practitioners to a superior authority. Once people believe they have calculated humanity's interests, everyone else's objections are just obstacles to their conceived optimal result. Give that conviction a financial base and effective altruists couuld spend generations deciding which problems matter and whose judgment should be taken seriously.

Nicholas Decker, a self-described effective altruist, laid out his case for political violence in his April 17, 2025 personal Substack essay, “When Must We Kill Them?” He argued that canceled elections, defiance of courts and other assaults on democracy would justify revolutionary violence against the Trump administration, while urging readers to wait and choose their own thresholds. “In short, we must decide when we must kill them,” he wrote. The authority he claims is extraordinary: he and his readers decide when their assessment of the government becomes permission to kill. The people choosing the threshold would also decide whether it had been crossed, with other people's lives riding on the answer.

Who gets to define the good?

The proposed arrangements would put money behind Anthropic’s vision of the public good while preserving its founders’ power over key company decisions. The Public Benefit Fund would be a wholly company-owned LLC, with governance details promised later. A separate Founder LLC, described in Reuters' reporting on the company's draft IPO filing, would direct a Class F share carrying 50.1% of voting power on key corporate matters. The fund would finance giving while the founder vehicle would preserve power over important company decisions.

Board control would be divided between Class F and Class A stockholders, who would elect three directors, and the Long-Term Benefit Trust, which would elect four. The trust is supposed to represent the broader public interest. Yet Anthropic's board selected its original trustees, and trustees elect their successors. The people affected by the company's decisions have no vote in choosing those representatives.

Anthropic's founders describe themselves as “distinctly equipped to be stewards of our mission,” according to the draft filing reviewed by Reuters. The company has already chosen the structure through which the public interest is represented; its chief executive now wants a role in setting the conditions under which rival builders may operate. That is a grandiose view of private authority: insiders define the mission, establish its guardians, and seek rules extending their judgment beyond their own company. The public is entitled to reject those priorities, choose other experts and accept different risks. Buying a Claude subscription gives Anthropic revenue, not a mandate to decide humanity's future.

The public also needs to see where the money goes. A private foundation's public Form 990-PF discloses grant recipients, amounts and leadership compensation. Anthropic's announcement contains no equivalent commitment to grant-by-grant reporting. Anthropic should be required to publish that information annually so citizens can trace which institutions it supports, how much they receive and what interests their work advances. Accountability cannot depend on the examples a company chooses to publicize.

A worldview with an endowment

At a hypothetical $2 trillion valuation, the proposed 5% stake would have a paper value of $100 billion, while the roughly 20% aggregate commitment would imply $400 billion across the separate arrangements. Reuters reported on October 15 that some investors were discussing $1.5 trillion, implying $75 billion and $300 billion respectively. These are potential equity values, dependent on approval and final terms, rather than an achieved IPO valuation or cash available to distribute.

Money on that scale could sustain a worldview through generations of political setbacks. Well-funded institutions can keep hiring researchers, cultivating experts and developing policy after an election defeat or a collapse in public enthusiasm. Their donors' priorities survive because the institutions advancing them can afford to wait. Over time, the power to finance years of patient work becomes the power to shape which arguments policymakers encounter and which experts they recognize.

Anthropic says none of the fund's Anthropic equity would support lobbying or political giving. But research agendas, professional authority and regulatory standards offer other routes to influence. Those routes matter especially when the company's preferred conception of safety becomes the basis for rules governing everyone else's ability to build.

When the mission becomes the rulebook

In his September essay, “We Must Pace the Frontier,” Dario Amodei sought a narrow antitrust waiver for safety discussions among AI companies and regulation reaching frontier labs unwilling to cooperate voluntarily. He favors capability-and-safety checkpoints, while also asking that limits on training compute, training runs and internal AI-assisted AI development be considered. Embedded evaluators would help verify compliance.

Anthropic is also asking Congress to narrow state scrutiny of its industry. Section 14 of the circulating discussion draft attributed to the company would displace specified state requirements for AI risk disclosures, third-party evaluations and incident reports, extending that protection even to developers outside the bill's federal frontier-model thresholds. After criticism, Anthropic announced plans on October 18 to revise the language. But the draft shows the bargain it put on the table: new federal duties in exchange for taking these oversight powers away from the states. A company claiming exceptional responsibility for the public good is helping write the limits on what the public can demand from it.

Under Amodei's September proposal, the firms already at the frontier would help shape standards constraining themselves and their competitors. If those standards become conditions for building, the people who define acceptable risk acquire leverage over who can challenge the incumbents. A safety judgment could become a barrier to entry, making evaluator independence a question of economic power as well as technical competence.

Anthropic's existing evaluator relationships show why that independence needs examination. On September 18, it announced that it would directly fund Accenture's evaluation work. Their December 2025 commercial agreement had already established a dedicated business group and planned Claude training for roughly 30,000 Accenture professionals. A firm helping sell and deploy Claude will also be paid to assess its safety. Anyone relying on those assessments should know how the commercial relationship is kept from influencing the work.

The nonprofit evaluator METR has six staff members with extensive internal access and the right to publish without an Anthropic veto. Anthropic nevertheless retains final authority over whether information falls into the agreed categories requiring redaction, such as confidential intellectual property. METR can publicly question those restrictions, but the company being evaluated still controls that boundary around the evidence readers can inspect.

Anthropic's evaluator agreements, which predate the new fund, show the accountability problem any larger network of privately financed oversight would have to solve. If evaluators' judgments help determine whether competitors can build, the public needs to know who pays them, what they can examine and what their contracts allow them to say.

Keep the gate open

I want powerful AI widely available and independent builders able to challenge the largest labs. That requires confronting demonstrated harms without allowing a company's predictions about the future to become an unchallengeable reason to restrict everyone else. The evidence must be open to scrutiny, the rules must apply consistently, and people outside the dominant safety networks must have a meaningful opportunity to contest both.

Anthropic should publish its grant recipients, conflicts, governance rules and results, and fund institutions free to prove it wrong. Congress should demand transparent, contestable standards and refuse to make the best-funded safety network the gatekeeper for the rest of the country. A philanthropic pledge gives its authors no special mandate over public policy.

The effective altruists are welcome to give away their fortunes, but humanity has not appointed them to govern its future.