Thinking about how organisations perform in an AI-enabled world

AyEye Today · 27 October 2026

Anthropic has divided control. Accountability still needs an address

Abstract painting of oxblood, turquoise and saffron-violet forces pressing around an off-centre raw-canvas void.
Original artwork: Unresolved Authority. Chiappe × OpenAI.

Anthropic’s S-1 maps a novel control system. Its real test is whether evidence can reach authority before consequences become irreversible.

The story, brieflyEvidence · Interpretation

The argument

Anthropic has split formal power among founders, a trust intended to operate independently, a public-benefit board and a fund. Accountability depends on the routes between them.

What is new

Founders receive 50.1 per cent of general voting power. The trust generally appoints a board majority; directors must balance investors, affected people and public purpose.

Why it matters

A model-release conflict can move faster than director terms or stockholder votes. Formal power matters only if contested evidence reaches a decision owner in time.

LecXie’s practical interpretation

Map six fields: trigger, owner, record, escalation, clock and review. Publish the route and standard even when underlying safety evidence must remain confidential.

Evidence & limits

The S-1 supports the control map and five-per-cent fund contribution. It is a securities document, not proof that internal escalation processes are absent.