There is something wonderfully modern about investing in a company that might end civilisation, then funding organisations that want governments to restrain it. Jaan Tallinn, Skype’s co-founder and an early Anthropic investor, occupies precisely this position. It invites suspicion. It also invites a little more reading than a conspiracy diagram usually accommodates.
Julia McCoy’s criticism of the AI-safety funding network raises a legitimate question: when the investor, the researchers and the regulation advocates share financial connections, whose interests are being served? The answer becomes interesting once we stop treating either philanthropy or self-interest as a complete explanation.
The financial connection is public. Anthropic announced that Tallinn led its $124 million financing round in 2021. The Survival and Flourishing Fund’s 2025 recommendations identify him as the source behind $2.035 million for Daniel Kokotajlo’s AI Futures Project, including a conditional $500,000 matching pledge. These are funding recommendations and commitments, not proof that every dollar arrived that year. Still, the overlap is real. Nobody needed night-vision goggles to uncover it.
Tallinn’s stated motivation has a longer history than his Anthropic investment. In an October 2026 interview with The Generalist, he traced his interest in AI risk to 2008 and his first safety donation to 2009. He said he asked Anthropic’s founders to obtain competing investment offers before accepting his money, to establish that he was replacing conventional investors rather than creating another frontier laboratory.
That is his account, not an independently verified counterfactual. But it matters: a longstanding concern is harder to explain as a story invented to protect a recent investment. He also says investment gives him influence and resources for safety work. The BBC recorded his explanation that investing buys access to conversations inside AI companies. He is quite open about wanting a seat near the kitchen.
His political ambitions are equally explicit. In a May 2026 Manifold interview, Tallinn described moving from supporting research and engaging legislators towards generating public demand for rules. He argued that advocates should say what genuinely worries them, rather than substitute more popular complaints about AI. Funding public persuasion is therefore part of his declared strategy. A secret agenda becomes rather less secret when its sponsor explains it on a podcast.
None of this cancels the conflict of interest. An investor can sincerely fear catastrophe and benefit from rules that favour a company he owns. Funding also helps determine which questions receive research budgets, which experts become available to journalists, and which policy options look respectable. It can shape the conversation without anyone receiving instructions about what to conclude.
Would the AI Futures Project’s proposals help Anthropic? Some mechanisms plausibly could. Audits, secure infrastructure and regulatory engagement impose costs that established companies may absorb more easily than challengers. Slowing new training while permitting existing products to operate could prolong the commercial life of models already developed. These are economic possibilities, not evidence of Tallinn’s intentions or a calculation of Anthropic’s eventual profits.
Anthropic’s own regulatory preferences offer a useful comparison. Its SB 53 compliance statement says many required practices evolved from procedures it already followed. Making familiar practices mandatory could give an experienced company an adjustment advantage. But the company also supports limiting obligations to large frontier developers, protecting smaller firms. And Anthropic’s preferred rules must not be casually substituted for the AI Futures Project’s programme.
That programme, “AI 2040: Plan A,” contains substantial obstacles to a simple story about protecting Anthropic. It recommends slowing the frontier so that dozens of companies across countries can catch up. Its preferred transparency regime would make almost all new algorithms public. Outsiders could purchase research compute and investigate frontier development. For a business whose advantage includes proprietary research, this is an awkward sort of favour.
There are important limits. Public algorithms do not mean freely downloadable model weights: the proposal protects those separately. Nor does access to research erase advantages in capital, distribution or infrastructure. Nevertheless, enabling competitors to learn and catch up could erode Anthropic’s lead. Delaying advanced systems could also postpone revenue. The net commercial effect cannot honestly be read straight off the donor list.
The implementation details deserve more suspicion than the slogans. One illustrative transition in the verification plan would temporarily shut down small compute clusters completely while allowing larger facilities to retain some operation. The authors justify this through verification logistics. Economically, it could disadvantage smaller independent operators. A plan intended to disperse power can still produce rules that favour those who already possess it.
The most credible concern is selective adoption. Governments might embrace costly verification, restricted infrastructure and approved operating environments while quietly losing enthusiasm for research transparency and meaningful access for competitors. Anthropic and other incumbents could benefit from that combination. This is a plausible failure mode, not a demonstrated arrangement. It requires scrutiny of the actual rules, their exemptions and who controls admission.
Tallinn’s public record is consistent with a sincere, persistent effort to prevent outcomes he considers disastrous, coupled with an explicit desire to influence policy. His financial position still warrants disclosure and scrutiny. Neither conclusion makes the other disappear.
The useful response is to demand published funding, independently challengeable safety claims, proportionate requirements and credible routes for new entrants. Ask who bears the costs, who gets access and which privileges expire. Civilisation may need a fire alarm. It should also check who owns the building, who writes the inspection rules, and whether the emergency exit opens for everyone.




No comments yet