Anthropic's $2 trillion IPO turns its trustee experiment into a public-market question

Anthropic is heading for an IPO that could value the Claude maker at $2 trillion — and the Long-Term Benefit Trust, the small group of self-appointed advisers who control a majority of Anthropic's board, is about to be examined by a much broader and less forgiving investor base. The trust has never had to draw a hard red line against Anthropic's commercial leadership; that is about to change.

The Long-Term Benefit Trust (LTBT) holds no equity but appoints or dismisses four of Anthropic's seven directors, including Netflix co-founder Reed Hastings and Novartis CEO Vas Narasimhan. The trust's three current members — chair Neil Buddy Shah of the Clinton Health Access Initiative, former Federal Reserve chair Ben Bernanke, and Richard Fontaine, CEO of the Center for a New American Security — meet weekly among themselves and with Anthropic's leadership every other week. They get advance notice of major actions including new model launches and have weighed in on the Mythos cybersecurity model's limited rollout via the Glasswing Project, as well as Anthropic's dispute with the US government over automated weapons.

So far the trust has operated largely in an advisory capacity. It has not forced a serious trade-off between profit and the company's public-benefit mission. Anthropic's trustees "meet weekly among themselves, while also meeting Anthropic's leadership as frequently as every other week," according to people close to the matter — but have yet to draw a red line on a commercial-versus-mission conflict. Harvard Law professor Jesse Fried argues the structure has a "built-in conflict" because it lets "self-appointed individuals decide how much profit to sacrifice for the firm's mission" while the company raises capital from profit-seeking investors. Anthropic's structure is widely seen as less risky than OpenAI's because the LTBT carries an 85 percent shareholder "kill switch" — but that supermajority could be renegotiated once the company is public.

The trust is being pitched as an industry blueprint on the order of GAAP accounting, with Bernanke's July appointment signaling an intent to build something more institutional. Whether it can constrain Anthropic's leadership when commercial pressure intensifies — and whether public-market investors will tolerate a governance layer that can ignore profit in decision-making — is now the real question. As one venture capitalist who backed Anthropic put it, "there was a judgment made by investors that capitalism would win in the end." The IPO will test whether that judgment survives contact with a trustee body explicitly designed to push back on it.

Resect launches with $25 million to "resect" hallucinations out of open models. Seattle-based Resect AI raised $25 million from undisclosed private equity backers to commercialize an internal-visibility stack that watches large language models run, detects when they are about to confabulate, and redirects them before they produce a wrong answer. The company's NeuroWave Product Suite — what CEO Kevin Owens calls "a polygraph for neural networks" — started as an internal tool Resect built while training its own factuality-focused models and now ships as an open-source add-on for DeepSeek, Qwen, and Llama. Owens argues hallucinations are an enterprise-trust problem, not just a research curiosity: "AI has prematurely been put in a position of trust… AI must be anchored in truth to be widely adopted across the enterprise." Funding is earmarked for R&D, go-to-market, and hiring in the Seattle and Portland regions.

Microsoft tells a judge that less than 1 percent of Copilot chats regurgitate copyrighted text. In a new filing in the consolidated Authors Guild v. OpenAI lawsuit, Microsoft argued that an analysis of more than 8 million chat logs found fewer than 1 percent regurgitated 16 or more words from copyrighted sources, and urged the court to treat AI training on copyrighted material as fair use. The filing lands weeks after the Trump administration filed a statement of interest backing OpenAI in the parallel New York Times v. OpenAI case. Microsoft argues that systems like Copilot are used for "significantly different purposes than the original," so even occasional reproduction "hardly undermines the transformative purpose of LLM training." The judge will weigh Microsoft's numbers against the Authors Guild's claim that training itself is the infringement, not the regurgitation rate.

What to watch: whether the LTBT's first hard call comes from the Mythos rollout, the AI-automated-weapons dispute, or a yet-unannounced model launch — and whether Microsoft can translate a sub-1-percent regurgitation rate into a fair-use win at the training stage.

Will public-market investors accept a board majority controlled by people with no equity stake — or will the trust's $2 trillion moment end with the same kind of reckoning OpenAI's board saw in 2023? Tell us in the comments.