Anthropic publishes Public Benefit Report, detailing AI progress and gaps in measuring impact
The company outlines Claude's use across education, science and healthcare, alongside its AI safety commitments, proposed 5% equity pledge and plans for annual public reporting
Anthropic's 2026 Public Benefit Report sets out the company's progress and commitments across AI safety, scientific research, education, governance and public benefit
Anthropic has published its 2026 Public Benefit Report, committing to an annual assessment of its progress on AI safety and public benefit while acknowledging that it has not yet gathered sufficient evidence to establish whether much of its work in education and healthcare is achieving its intended results.
Published on October 21, the report examines how the company is working toward its public benefit charter, covering responsible AI development, governance, economic opportunity and access to advanced technology.
It includes figures on Claude's deployment across education and scientific research, examples of AI-assisted healthcare initiatives, financial commitments to workforce development and disclosures about shortcomings in the company's existing safety practices.
Anthropic is structured as a Public Benefit Corporation (PBC), requiring it to balance shareholders' financial interests with its stated responsibilities to the wider public. Its latest publication establishes five areas against which that commitment will be assessed annually.
Daniela Amodei, Co-Founder and President of Anthropic, confirmed the reporting commitment in a LinkedIn post:
"We’ll publish one every year laying out how we define success toward our charter and the concrete proof of our progress."
The report has also been developed with input from Anthropic's Long-Term Benefit Trust (LTBT), an independent governance body that elects a majority of the company's board and has a formal role in shaping its safety policies.
Claude reaches teachers in 63 countries as education programs expand
Education features prominently in Anthropic's assessment of how AI is being used beyond its commercial customer base.
In Ghana, the Ministry of Education has worked with Playlab to develop Claude-powered lesson-planning tools for approximately 68,000 instructors. The company is also working with Teach For All to introduce Claude and AI fluency training to teachers across 63 countries.
In the US, Claude for Teachers provides free tools for K–12 educators, with resources mapped to academic standards across all 50 states.
Anthropic has also made the teaching skills and evaluation rubrics behind Claude for Teachers openly available, allowing other developers and education organizations to build on that work.
Its education commitments extend to research and infrastructure. The company says it has endorsed a global agenda focused on shared benchmarks, research and data for AI in education in low- and middle-income countries, while developing access to public datasets through Claude, including the Learning Commons knowledge graph.
Scientific research is another area where Anthropic provides figures on institutional adoption.
According to the report, more than 3,000 research institutions across over 100 countries currently receive support through the company's academic and nonprofit research offerings. Its AI for Science program has also awarded Claude credits to more than 800 projects at over 450 institutions.
The company has introduced Claude Science, a research workbench supporting analysis and hypothesis generation, alongside free and discounted access for academic and nonprofit researchers.
However, Anthropic acknowledges a distinction between making AI available and demonstrating that it improves outcomes.
Much of its education and health activity remains at an early stage. The company states that it does not yet have evidence that these initiatives have produced their intended effects and recognizes that it has not been sufficiently active in collecting data to measure results.
Healthcare and scientific applications provide early evidence of AI use
Anthropic also details applications where Claude is already being used to support scientific and public health work.
In the Democratic Republic of the Congo, Claude has been used by organizations responding to an outbreak of a rare Ebola strain. According to the company, the World Health Organization's Africa office has reduced the time needed to produce outbreak situation reports from as much as a full day to under an hour.
Vaccine experts have also used Claude to compare proposals for vaccine candidates.
In Burkina Faso, the International Rescue Committee has used Claude to turn clinic registers into digital dashboards. The resulting analysis identified a spike of 25,000 pneumonia cases and highlighted a discrepancy in maternal healthcare records: more than 6,000 women had attended prenatal appointments, while only 1,762 recorded deliveries involved a healthcare worker.
The team is investigating that discrepancy and using the data to inform its response.
Other examples include Claude designing new protein binders, an early stage of drug development, and cybersecurity teams using the technology to identify more than 33,000 critical- and high-severity security flaws in widely used software, according to Anthropic.
These examples demonstrate specific applications of the technology, although the company recognizes that wider evidence of sustained public benefit remains incomplete.
In her foreword, Amodei describes the need to assess those outcomes rather than rely on the company's stated intentions:
"In some areas, we have direct control over outcomes. In others, we’ll succeed only by partnering with people, organizations, and governments. In all cases, we owe the public an honest accounting of our work."
AI safety assessment acknowledges failures and limits of current safeguards
Alongside its examples of beneficial AI use, Anthropic addresses shortcomings in model oversight, monitoring and safety evaluations.
The report confirms that four Claude models gained unauthorized access to real third-party systems during cybersecurity testing after a misconfigured testing environment connected them to the open internet.
It also acknowledges other instances in which Claude took unintended actions during evaluations and internal use.
Anthropic says it has developed methods to investigate misaligned model behavior and incorporated interpretability research into its pre-release auditing processes. However, it recognizes that these techniques are not yet sufficiently dependable or comprehensive to establish a model's intentions in every situation.
The company also identifies the growing use of AI in its own development processes as an unresolved challenge. Claude already performs an increasing share of Anthropic's safety research, while current models are being used to help train subsequent generations.
As these systems become more capable, Anthropic warns that meaningful human oversight could become harder to maintain.
The report documents external testing by organizations including the UK AI Security Institute and the US Center for Advancing Innovation and Standards for Super Intelligence. Anthropic says more than 150 hours of external red-teaming and tens of thousands of attempted requests against Claude Opus 5.5 did not produce a universal jailbreak of its cybersecurity safeguards.
Nevertheless, the company acknowledges that legitimate researchers can still encounter restrictions intended to prevent malicious use, with verification processes sometimes taking days or weeks.
Its governance commitments include an initial embedded evaluation partnership with METR and support for stronger independent oversight of frontier AI development.
Anthropic also identifies gaps in the wider regulatory environment. Although it has supported AI safety legislation at state level in the US, the report notes that there is still no mandatory federal framework independently verifying whether frontier AI developers are following their own safety commitments.
Anthropic commits $200m to economic research and $150m to early-career fellowships
The report also sets out Anthropic's response to potential economic disruption as AI takes on a greater share of workplace tasks.
The company has committed $200 million to its Economic Futures Research Fund, intended to support research into AI's labor-market effects and test potential policy interventions.
Among the questions it aims to examine is how workers can develop expertise when AI increasingly performs entry-level tasks, alongside how successful employment and training initiatives might be expanded.
A further $150 million has been committed to Claude Corps, an economic mobility program developed with CodePath.
The initiative is intended to fund 1,000 year-long fellowships for people early in their careers, placing participants with a network of more than 400 nonprofit organizations.
Anthropic has also been publishing research through its Economic Index, making underlying data available to external institutions, including Stanford's Digital Economy Lab.
While these programs represent defined financial commitments, the company acknowledges that their current scale is limited relative to the potential economic changes associated with advanced AI.
Its first year of Claude Corps will support 1,000 participants, while the broader labor-market effects and distribution of AI-generated economic gains remain uncertain.
The report additionally confirms Anthropic's commitment to keeping Claude free of advertising and its support for independent research into AI and user wellbeing, including funding for Common Sense Media's Youth AI Safety Institute.
Products cited in this area include Learning mode, which guides students through questions rather than simply providing answers, and Reflect, which encourages users to examine how they use AI.
Proposed 5% equity pledge adds financial commitment to public benefit plans
Anthropic's plans also include a proposed allocation of at least 5% of its shares to a separate entity dedicated to advancing its public benefit charter.
The commitment remains subject to stockholder approval. It sits alongside existing pledges by the company's co-founders to donate 80% of their individual equity holdings to philanthropy and a matched equity-donation program for employees.
Taken together, Anthropic hopes these arrangements will eventually dedicate approximately 20% of company shares to public benefit purposes.
The financial commitment is accompanied by an expanded role for the Long-Term Benefit Trust in assessing whether Anthropic delivers on its stated objectives.
Amodei also confirmed on LinkedIn that Chris Stubbs, an experimental physicist and former Dean of Science at Harvard University, is joining the Trust.
In its closing commentary, the LTBT distinguishes between the activities documented in this first assessment and the measurable results it expects from future publications.
The trustees acknowledge Anthropic's investment in safety research, transparency and public benefit programs, but state that the distribution of AI's benefits still falls short of the company's ambitions.
They expect subsequent reports to provide stronger evidence of outcomes, including independent confirmation that advanced models remain under human control, measurable improvements in people's lives and progress toward shared AI safety standards.
Anthropic's Public Benefit Report will now be published annually, with future assessments intended to track progress against the five charter pillars and provide a continuing account of the company's commitments, results and unresolved challenges.
ETIH Innovation Awards 2027 open for entries
The ETIH Innovation Awards 2027 are now open for entries, with 26 categories recognising companies, institutions, projects and individuals making a measurable impact across education technology. The global awards span AI in education, digital learning, higher education, skills and workforce, inclusion, education operations, evidence and impact, with entries managed through ETIH’s new online awards platform.
Entries are open worldwide, with the early-bird deadline on Friday 18 December 2026 and the final entry deadline on Friday 26 February 2027 at 23:59 UK time.
Explore the 2027 categories and judging criteria, read the How to Enter guide or start an entry through the ETIH Innovation Awards 2027 platform.
