← Back to all entries
2026-02-26 ✅ Best Practices

Dario Amodei's Open Letter on AI Responsibility & February Risk Report PDF

Dario Amodei's Open Letter on AI Responsibility & February Risk Report PDF — visual for 2026-02-26

Dario Amodei — Open Letter on Frontier Lab Responsibility

Anthropic CEO Dario Amodei has published a personal open letter titled "What We Owe Each Other in the Age of Transformative AI" on the Anthropic blog. The letter is addressed to the broader AI research and technology community and articulates what Amodei frames as the mutual obligations of organisations building frontier AI systems — to each other, to their users, and to society at large. It is notably personal in tone compared to Anthropic's usual policy communications.

The letter's central argument is that the current period — characterised by rapid capability gains, intense competitive pressure, and inadequate governance infrastructure — requires frontier labs to exercise a higher standard of self-restraint than market incentives would naturally produce. Amodei identifies three specific obligations he believes all frontier labs share: maintaining credible transparency about capabilities and risks, engaging constructively with regulatory processes even when those processes are imperfect, and refusing to deploy systems whose risk profile is not yet understood.

The letter is being widely discussed in the AI policy community as one of the most direct public statements from a frontier lab CEO about the collective-action dimension of AI safety. Responses from other lab leadership are anticipated in the coming days.

Dario Amodei safety policy Anthropic AI governance retrospective

February 2026 Risk Report — Full PDF Now Available

The February 2026 Risk Report, which was first announced and summarised on February 15, has today been released as a formatted PDF document via Anthropic's research portal. The PDF edition includes additional appendices that were not published in the initial summary: a full methodology section for the ARARA evaluation cluster, detailed scoring rubrics for the revised CBRN uplift thresholds, and a technical glossary covering the new terminology introduced in RSP v3.0.

The document is intended for use by policymakers, academic researchers, and enterprise compliance teams who require a citeable, versioned reference for Anthropic's safety evaluation methodology. Anthropic has assigned the document a formal citation handle and DOI to support academic use. The PDF is available without login at anthropic.com/research.

risk report safety research Anthropic retrospective