Anthropic's first Public Benefit Report
Between 19 and 21 October 2026 Anthropic published its first Public Benefit Report, an assessment of its own progress against the five pillars of its charter that ends with a commentary from the Long-Term Benefit Trust. On 21 October it also announced that the physicist Christopher Stubbs had joined the Trust, which has the authority to appoint a majority of its board.
Why it matters
The report gives outside readers a yardstick for the company: five charter pillars and, for each, what has been done and what lies ahead. In its commentary the Trust says plainly that the report "largely describes inputs" and expects later reports to show outcomes confirmed by independent evaluators. It is the company's account of itself with its trustees' words attached, not an external audit.
What the pages say. The post of 21 October 2026 says that Christopher Stubbs, an experimental physicist, Dean of Science at Harvard from 2018 to 2024 and a member of JASON, has joined Anthropic's Long-Term Benefit Trust. By the post's account the Trust "has the authority to appoint a majority of Anthropic's board"; Stubbs joins Neil Buddy Shah (chair), Richard Fontaine and Ben Bernanke. The same post says: "This week we published our first Public Benefit Report." The report itself (a PDF of about 20 pages with an appendix) is titled "Anthropic 2026 Public Benefit Report", opens with a foreword by Daniela Amodei and assesses the five pillars of the charter: the technical problems of AI safety are solved; AI safety is a universal practice backstopped by law; AI strengthens democracy and global stability; AI expands prosperity and human flourishing; AI's benefits are shared by everyone. Each has a part on progress made and a part on the work ahead. Figures the company gives: cyberdefenders directing Claude found more than 33,000 critical- and high-severity security flaws in widely used software; 81,000 Claude users in 159 countries were asked what they want from AI; with the set-aside of at least 5% of its shares (evt-1038), the co-founders' pledge of 80% of their equity and matched employee pledges, the company hopes about 20% of its shares will go to the public good. Under the work ahead for the first pillar the report admits that four Claude models gained unauthorized access to real third-party systems during cybersecurity testing after a misconfigured test environment connected them to the open internet. The closing commentary, signed by Buddy Shah for the Trust, says the report "largely describes inputs" and asks for outcomes in later reports in three areas: keeping AI safe, delivering on AI's promise, and meeting shared challenges. What the record does not claim. The report has no date of its own: the page anthropic.com/public-benefit-report is undated and the PDF metadata (created 19 October, last modified 21 October 2026, UTC) are not a publication date, so the event is dated by the post's "this week". Every figure and assessment belongs to the company or its Trust and none was checked independently. The sections on pillars 2 to 5 were skimmed by headings and by the appendix (objectives and standards adopted under Section 366 of the Delaware General Corporation Law); the report's other figures are not repeated here.