Anthropic Pledges Shares to the Public and Pitches a High-Bar Safety Bill

Anthropic says it wants to set aside 5 percent of its shares for a new Public Benefit Fund, with the structure laid out in a blog post. The company still needs shareholder approval. And the stake is designed to rise with the company: for every 50 percent increase in Anthropic’s value above a baseline it has not disclosed, the board can add another 2 percent.

At the valuation Anthropic’s backers are aiming for in a planned I.P.O. — more than $2 trillion — that 5 percent would be worth about $100 billion. The fund could back large donations to nonprofits or direct payments to specific groups of people. Anthropic points to the labor market as one possible target, which is an awkward area for a company already reshaping entry-level work: junior jobs are shrinking at Anthropic.

The company says it has already committed $200 million to research on economic futures, but adds that “bolder approaches” may be needed. The fund would not be allowed to use its shares for lobbying or political donations. Anthropic says that, together with earlier commitments, about 20 percent of its shares are now dedicated to the public good.

That total includes the co-founders’ pledge to donate 80 percent of their equity and a program that matches employees’ share donations. Anthropic says it will publish more detail on governance and valuation milestones in the coming months. Elon Musk, whose xAI is in the same race, shot back on X that he recommends giving the shares directly to the public and said, “This is what I will be doing,” without saying how or when.

On regulation, Anthropic is taking a much tighter line. Politico reported that the company has shown congressional staff a draft A.I. safety bill that would apply only to frontier-model developers trained with more than 10²⁵ computing operations and either more than $500 million in annual A.I. revenue or more than $1 billion in A.I. research spending. Those firms would have to publish safety frameworks, file transparency reports, allow independent evaluations, report incidents to the Commerce Department within 72 hours and protect whistle-blowers.

But the stop-power in the draft is narrow. Only the Commerce Department could halt a model, and only by going through federal court. The standard would be a foreseeable, imminent risk of more than $1 billion in damage or more than 50 people killed or seriously injured. The draft names biological weapons, large-scale cyberattacks and models that evade their creators’ safety controls as the main dangers. Anthropic says the proposal matches its “Advanced AI Framework” and would replace a messy patchwork of state laws, while critics say it leaves too much power with one agency and may favor big labs over smaller rivals. The bill has not been formally introduced.