When an AI company goes public, safety takes a back seat. That's the playbook. Every quarter, earnings calls get flooded with analyst questions like "why is your model slower than the competition?" — and eventually, speed and profit win.
Anthropic decided not to give its board to shareholders.
Here's what everyone believes — "going public means selling out safety"
There's real logic here. Public markets demand quarterly results, and the moment OpenAI ships a faster model, analysts on Anthropic's earnings call will publicly ask whether the company is "being too conservative." Stack enough of those questions and CEOs eventually move.
The numbers make the pressure feel concrete. Annualized revenue run-rate hit $47B in May 2026, with projected Q2 2026 operating income of ~$559M. But compute costs alone run about $1.25B a month. That operating profit is temporary — Anthropic itself disclosed that margins disappear again in the second half of 2026.
The structural pressure public markets create
Anthropic currently puts nearly all revenue back into R&D — compared to Google's ~15%. In public markets, that ratio reads as "inefficient." Post-IPO, investor questions about profitability will start immediately, turning the "why not maximize returns?" pressure into a recurring quarterly feature.
But the data says the opposite — the board escaped shareholder control
In a typical public company, the board is elected by shareholder vote. That's how activist hedge funds buy in, install directors, and push the quarterly pressure lever. Anthropic blocked this path at the design stage.
The first line of defense is the Public Benefit Corporation (PBC) structure. A standard corporation has a legal obligation to maximize shareholder returns — prioritizing safety over profit can get directors sued for breach of fiduciary duty. A PBC enshrines public benefit alongside shareholder interests in its charter. Safety-first decisions that reduce profits come with legal immunity.
The second is the Long-Term Benefit Trust (LTBT). Founded in 2023, this body is made up of five outside individuals who own not a single share of the company. They hold Class T stock that gives them power to elect and replace board directors — and as of April 2026, they already control the board majority. It doesn't matter how much Amazon or Google have invested. They can't take over the board.
| Typical Public Company | Anthropic (PBC+LTBT) | |
|---|---|---|
| Board election | Shareholder vote | LTBT external trust (zero shares) |
| Safety-first decision protection | None | Legally protected by PBC charter |
| Activist hedge fund director seats | Possible | Structurally blocked |
| Post-IPO safety compromise path | Earnings call → CEO pressure | LTBT board as buffer |
Here's the thing though — the original designers are all gone
The LTBT structure being complete doesn't end the story. We need to talk about the people who built it.
The original five trustees at LTBT's founding in 2023: Jason Matheny (RAND CEO), Paul Christiano (Alignment Research Center founder), Kanika Bahl (Evidence Action CEO), Zach Robinson (Effective Ventures CEO), and Neil Buddy Shah (Clinton Health Access Initiative CEO, Chair). A lineup built around AI safety researchers and social enterprise leaders.
By January 2026 — five months before the IPO filing — Matheny, Christiano, Bahl, and Robinson had all left. Current trustees include Neil Shah (Chair), Richard Fontaine (CNAS CEO), former California Supreme Court Justice Mariano-Florentino Cuéllar, and two undisclosed seats.
The trustee composition shifted from AI safety researchers to national security and legal experts. Whether that's a strengthening or weakening of the original intent — the decisions ahead will tell us.
The first real test already happened. In February 2026, the Pentagon placed Anthropic on its supply chain risk list. Reason: Anthropic refused to allow Claude to be used for mass surveillance and fully autonomous weapons. Anthropic accepted the commercial cost of losing government contracts to hold its usage limits. The blacklisting was lifted 19 days later — but the precedent is set: safety-first decisions come with real costs.
"The company that files first gets to define how a frontier AI lab reports its financials in public markets — a structural advantage that could shape how investors value the entire sector."
— Gil Luria, DA Davidson
How to actually judge this governance experiment
Whether you're an enterprise running on Anthropic's APIs or considering an investment — there are concrete checkpoints to assess whether this governance structure actually holds. Track these four after the IPO.
- Watch who fills the undisclosed LTBT seats
Two of five seats are currently undisclosed. Post-IPO S-1 disclosures and annual 10-K filings will reveal these individuals. If AI safety and ethics expertise drops in the composition, that's signal one. - Monitor board composition changes post-earnings
Watch which directors the LTBT installs after earnings calls. A shift toward profitability-focused profiles indicates the design intent is drifting. - Track safety decisions that carry a commercial price
Watch whether cases like the Pentagon blacklisting — where safety priorities meant lost contracts — keep happening. If these decisions stop appearing, the structure may be yielding to pressure. - Frame the first two earnings calls
Watch whether analysts frame Anthropic's safety-first policies as a "risk" or as a "competitive moat." That market interpretation shapes the CEO pressure loop that follows.
The structure matters less than the people inside it
Even with the LTBT's legal protections in place, if the trustees themselves decide "growth matters more," nothing stops them. The real variable in this experiment isn't the institution — it's the humans inside it.
