Beacon Papers No. 20 — AI Safety Cannot Run on Promises

Nov 16, 2026 | Papers & Reports, News, Publications

A principle tells power where to stop. Who shows whether it stopped? In Beacon Papers No. 20, Nguyen Anh Tuan connects the work of the November 10 BGF Conference with a practical architecture for AI safety: a Constitution, an Observatory and a Movement.

The Constitution defines limits on power. The AIWS Trust Observatory examines whether commitments are kept. The Human–AI Co-Creation Movement sustains people’s ability to understand, question and shape the technologies they use, with the Atlas providing a record of their contributions.

The paper calls this wider responsibility civilizational safety: humanity’s capacity to know what AI power is doing, judge its consequences and correct its course. It proposes public disclosure, independent observation and distributed responsibility, holding the institutions themselves to the standards they ask of others.

The next task is to make safety claims observable, contestable and answerable to the people whose lives they affect.

Read and download the full text of Beacon Papers No. 20 — “AI Safety Cannot Run on Promises” here.