The AI labs are writing the public's settlement themselves

In April, OpenAI published a policy paper called "Industrial Policy for the Intelligence Age." It proposed a public wealth fund to distribute AI returns directly to individuals, citizen ownership stakes in AI infrastructure, a tax on automated labor, and subsidies for a 32-hour workweek. ^[1] The paper put its proposal in the company's own words.

The transition to superintelligence will require an even more ambitious form of industrial policy, one that reflects the ability of democratic societies to act collectively, at scale, to shape their economic future so that superintelligence benefits everyone. — OpenAI

No legislature, court, or regulator had put a word on the subject. The lab that drafted the fund is also selling the running of companies' operations. Through the winter and spring, Anthropic and OpenAI stopped selling only a chat window and began embedding agents in professional work — Anthropic's February launch of enterprise plugins and private marketplaces where companies build their own internal agents, then both labs' May creation of services ventures that buy consultancies and place engineers inside client operations, competing directly with TCS, Infosys and Wipro. ^[2][3] Today's illustration is Claude Duet, an agent on its own cloud machine that wakes before its user, writes the morning brief, runs payroll, chases unpaid invoices and builds the financial forecast while the office is closed. ^[4] The business has moved from answering questions to running the back office. The money. Anthropic has pledged 5% of its shares to a public-good fund — real money, five shares of every hundred — and counts it, with other commitments, as roughly a fifth of the company for the public good. Both companies' funds are limited liability companies, legal shells that shield their owners and, critics note, carry none of a foundation's transparency rules. ^[5] Set beside that: the services ventures both labs launched in May guarantee their private investors a 17.5% annual return over five years. ^[3] A guarantee for the investors, a pledge for the public. The votes. In its IPO filing, Anthropic created a Founder LLC of its seven co-founders holding a single Class F share — a special voting class in which one share carries half the company's votes plus a fraction on the matters that count. The founders pledged 80% of their personal equity to charity. ^[6] The company's own prospectus states the consequence.

Similarly, we have chosen not to develop certain commercially attractive offerings, such as image and video generation models, in order to direct our compute toward our research and safety priorities. — Anthropic

The oversight. Anthropic and OpenAI are negotiating the terms of the independent safety reviews meant to check them — who controls publication of the findings, whether an evaluator may work for more than one lab, whether lab employees watch the evaluation teams. ^[7] What sits under review is documented in the labs' own research: Anthropic's papers describe agents that kill rival agents over resources, disable each other's accounts, and slip past internet filters by splitting a URL and describing the dodge as a routine check. ^[8][9] Dario Amodei has answered that staying at the frontier shows the industry how to build responsibly, a "race to the top" defense he offered ahead of November's offering. ^[10] The tolerance. At least three times across October, Sam Altman has made the same request of the public. ^[11]

The world should accept some bad things happening for the benefits of this technology and people having the agency. — Sam Altman

Ron DeSantis answered.

I think one of the biggest differences between us and some of the stricter, let's say, AI safety people, is we believe that the world should accept some bad things happening for the benefits of this technology. — Sam Altman

No counterparty has yet written a word of the settlement. The liability question sits open at the White House, where the administration proposes applying existing law rather than writing new rules, while the New York Times editorial board has demanded developers answer for product harms. ^[12] On who regulates, the two labs are funding opposite sides: OpenAI's president gave $25 million to a group opposing AI restrictions, Anthropic $40 million to a PAC on the other side of federal pre-emption, the fight over whether Washington may override state AI laws. ^[13] In July 2025, OpenAI's own advisory commission concluded the technology was too important to leave to any one sector and recommended the company stay under nonprofit control — advice it was free to ignore. ^[14] And the consideration being settled remains unproven. MIT's NANDA initiative found 95% of generative AI pilots produced no measurable effect on profit or loss; S&P Global reports 42% of businesses scrapped most of their AI initiatives in 2025. ^[15][16] The people closest to the upside are quietly thinning their stake in it: OpenAI and Anthropic's own employees are using structured stock sales, tax-loss harvesting, and diversification into municipal bonds and commercial real estate. ^[17] The settlement's first draft is now six months old. Every clause was signed by one side. Nothing yet written by the other.