The world does not need centralized control of AI. It needs independent global observation.
On September 23, 2026, an extraordinary warning came before the United Nations Security Council. Anthropic CEO Dario Amodei said that, if managed poorly, AI “could be a risk to humanity as a whole.” OpenAI CEO Sam Altman argued that the most important decisions about AI cannot be made by labs in San Francisco alone. Yet the United States rejected a global regulatory scheme.
This exposes the defining institutional problem of the AI Age. Frontier AI is becoming too consequential to remain visible only to those who build it, yet placing it under a centralized international authority could create another form of unaccountable power.
Beacon Papers No. 13 offers a third path: Do not centralize control over AI. Do not leave consequential AI power unobserved.
The paper introduces the AIWS Trust Observatory, established by the Boston Global Forum and AIWS as an independent civil-society institution that observes consequential AI power, verifies evidence, warns of emerging danger, and calls upon those with authority to act.
The need is urgent. Frontier laboratories can see what their systems are becoming. Society often sees only what they choose to disclose, or what surfaces after a failure. Self-disclosure is necessary. It is not independent oversight.
The Observatory examines not only what AI systems can do, but what authority they are given, whether safeguards are proportionate to that authority, and who must answer when harm occurs. It works through:
- Five Stations observing behavior, consequences, footprints, voluntary access, and protected insider evidence;
- a Commitment Register tracking whether companies and governments act on their public safety promises;
- four alert levels: Watch, Concern, Urgent Warning, and Critical Alert;
- the Frontier Capability Registry, the AI Incident Exchange, quarterly and annual reports, and a Blind Spot Map of what remains hidden from society.
The Observatory will not regulate companies or replace governments. Its authority comes from verified evidence, public warning, and the ability to mobilize those responsible. No one it observes may finance it, every finding carries an evidence grade, and it holds itself to the answerability it asks of others.
It is an institution designed to see danger before society meets it as catastrophe, and to scrutinize concentrated technological power without becoming a new center of unchecked power.
Humanity must not learn what frontier AI has become only after something irreversible has happened. AI power must be observed before it can be governed — and answerable before it can be trusted.
→ Read and download Beacon Papers No. 13: AI Power Must Not Operate Unseen
Dario Amodei addresses the United Nations security council in New York City on Wednesday. Photograph: Brendan McDermid/Reuters