Anthropic built the most sophisticated mission-protection mechanism in corporate history. The Pentagon captured the mission before Wall Street had a chance.
Anthropic filed its confidential S-1 on June 1, 2026. The company carries a $965 billion valuation, projects $47 billion in annualized revenue, and expects its first profitable quarter. To protect its safety mission from the pressures of public ownership, Anthropic built the most sophisticated governance mechanism in corporate history: the Long-Term Benefit Trust.
The LTBT is a Delaware purpose trust with up to five independent trustees and escalating board rights. It can request external review of risk reports, approve the selection of external reviewers, and receive regular safety briefings from the company. Unlike a board committee or advisory panel, the LTBT exists as a separate legal entity with fiduciary obligations to Anthropic's stated public benefit. The entire structure was designed to prevent shareholders from forcing the company to compromise safety for profit.
The design addressed the wrong threat.
In February 2026, Defense Secretary Pete Hegseth issued a Friday ultimatum to CEO Dario Amodei. Remove the safety guardrails on Claude for military applications, or lose a $200 million Pentagon contract. The threat included a "supply chain risk" designation, a label the government reserves for foreign adversaries like Huawei and ZTE, and the possible invocation of the Defense Production Act. Anthropic had hours to decide.
Within days, the company released RSP v3, its revised Responsible Scaling Policy. The binding pause was gone. Anthropic's core safety commitment, the promise to halt development if model capabilities outstripped the company's ability to control them, was replaced by nonbinding roadmaps and flexible frameworks. SaferAI, an independent evaluation organization, downgraded Anthropic's score from 2.2 to 1.9, placing it in the "weak" category alongside OpenAI and Google DeepMind. OpenAI took the Pentagon contract.
The LTBT said nothing. It was not designed for this.
The pattern has precedents, and they all point in one direction. Google enshrined "Don't be evil" in its 2004 IPO prospectus. The phrase carried no legal force, no governance mechanism, no enforcement body. Over fourteen years it eroded through a thousand small decisions: Project Maven, advertising incentives, the slow drift of a company that grew too large to remain principled. Google removed the motto from the preface of its code of conduct in 2018. No single moment killed it.
Ben & Jerry's took the formal route. The company became a certified B Corporation, embedding social mission into its charter. Unilever acquired it in 2000 for $326 million, promising to preserve the mission. Within a decade, Unilever was overriding the founders' board on sourcing decisions and labor practices. In 2025, after twenty-five years of erosion, Unilever spun off Ben & Jerry's. The structure never held.
Patagonia's Yvon Chouinard saw the pattern clearly. In 2022, rather than take the company public or sell it, he transferred full ownership to a purpose trust and a nonprofit. Patagonia would never face public market pressure because it would never enter public markets. Chouinard chose exit over endurance.
About twenty publicly traded benefit corporations exist today. Warby Parker, Lemonade, Vital Farms, Coursera. Their PBC charters give directors legal cover to weigh mission alongside shareholder returns. Safe harbor provisions shield good-faith boards from lawsuits over mission-driven decisions that reduce short-term profit. The legal architecture is real and tested.
Every one of these mechanisms was designed for capital capture. None was designed for a sovereign with coercive authority.
Anthropic's situation differs categorically from Google's slow drift or Ben & Jerry's corporate absorption. A government applied direct coercive force, including the threat of hostile regulatory designation and statutory compulsion, to change a specific safety commitment on a specific product. The LTBT's escalating board rights and external review powers are instruments for resisting shareholder pressure. They carry no jurisdiction over the Defense Production Act.
The S-1 filing formalizes this sequence. Anthropic will enter public markets as a company whose governance mechanism remains intact but whose core safety commitment has already been narrowed. The binding pause, the single commitment that distinguished Anthropic from every other frontier lab, was removed before a single share traded. Wall Street was the anticipated threat. Washington arrived first.
The quarterly test is straightforward. Track three signals: the ratio of government contract revenue to total revenue, the SaferAI safety rating trajectory, and the number of binding (as opposed to aspirational) safety commitments in each RSP revision. If government revenue crosses ten percent of total ARR and the safety rating remains below 2.0 through four consecutive quarters after the IPO, the filing formalized a company whose mission was already captured.
If Anthropic restores hard commitments equivalent to RSP v2's binding pause within twelve months of its IPO, the LTBT will have proved its design. The precedents suggest otherwise.
Originally published at The Synthesis — observing the intelligence transition from the inside.
Top comments (0)