A $100 Million Bet That the Discovery Was Never the Problem: Inside the OpenAI Foundation's Hepatitis C Initiative

Hepatitis C has been curable since 2014. An eight-to-twelve-week course of oral direct-acting antivirals cures at least 95 percent of patients. There is no research frontier standing between a person with hepatitis C and a cure.

And yet, across 2013 to 2022, only 34 percent of Americans with an initial hepatitis C infection were classified as cured or cleared. The state-level median was 29 percent. The spread ran from 10 percent in West Virginia to 51 percent in Connecticut. Federal elimination targets called for 58 percent clearance by 2025 and 80 percent by 2030; the country is not close to either.

That gap — between a drug that works and a population that never receives it — is the entire premise of the Breakthroughs to Follow-Through initiative, announced August 13, 2026, with $100 million from the OpenAI Foundation flowing through the Common Health Coalition. The stated goal is to at least double hepatitis C cure rates in eight states and localities within two years, then extend the same machinery to HIV prevention and curable cancers.

Why the money is going to logistics instead of science

Most large health philanthropy funds discovery. B2F funds the unglamorous work that happens after discovery: finding the people who tested positive and never started treatment, re-engaging patients who fell out of care, reducing the administrative load on clinicians and health departments, and coordinating between systems that do not talk to each other.

Dave Chokshi, who chairs the Common Health Coalition, described the tools jurisdictions are likely to want as ranging "from identifying people overdue for screening or treatment, to outreach platforms for re-engaging patients." That is a data-matching and case-management problem, not a biomedical one. It is also precisely the kind of work that has historically been unfundable — too operational for research funders, too technical for general operating support, and too unsexy for the kind of philanthropy that likes ribbon cuttings.

The Common Health Coalition is a plausible home for it. Founded by AHIP, the Alliance of Community Health Plans, the American Hospital Association, the American Medical Association and Kaiser Permanente, and formally announced in November 2023, the coalition was built to translate pandemic-era lessons into durable connective tissue between health care delivery and public health. Its four stated priority areas — coordination, always-on emergency preparedness, real-time disease detection, and exchange of actionable data — describe the hepatitis C cascade problem almost exactly.

Anna Makanju, who leads AI for Civil Society and Philanthropy at the OpenAI Foundation, framed the grantmaking rationale in terms grant seekers should note: "The organizations closest to society's greatest challenges have the deepest understanding of what communities need," but frequently lack the resources to adopt the technology that would help.

Four states first, four more within six months

Initial funds go to local collaborations and nonprofit organizations in Alabama, Illinois, Louisiana and Massachusetts, with additional jurisdictions to be named within six months to reach eight total.

The selection is not random, and understanding the logic tells you something about what a competitive application looks like. Alabama and Louisiana sit in the part of the country where clearance rates trail worst and where Medicaid coverage and rural access constraints compound the cascade. Massachusetts sits at the other end, with comparatively strong infrastructure — a place where doubling the cure rate means solving the hard residual cases rather than the easy first tranche. Illinois brings a dense urban jurisdiction with a large safety-net system. The portfolio is designed to produce evidence that generalizes, not just numbers that look good.

Two structural features of the money matter more than the headline number.

The grants are technology-neutral. Reporting on the announcement is explicit that grants "aren't tied to one AI platform or system provider." A funder underwritten by an AI company is not requiring grantees to use that company's products. For a nonprofit or health department, this removes the usual objection to corporate-adjacent technology philanthropy — that accepting it means accepting a vendor lock. Build the case for the tool that fits your workflow, not the tool that flatters the funder.

Technical assistance comes with the check. B2F explicitly provides support for the transition from pilot to full deployment. Anyone who has run a health-technology pilot knows this is where the failure lives: a successful six-month demonstration with a champion clinician, a promising evaluation, and then no path to production because nobody funded the integration, the training, or the sustainment.

What a fundable application actually looks like

If your organization operates in one of the four named jurisdictions — or in a state you believe could be among the next four — the work to do now is documentary, and it starts with your own cascade.

Quantify your local gap with real denominators. Not "we serve a high-risk population," but: how many people did you test in the last twenty-four months, how many were viremic, how many started direct-acting antivirals, how many achieved sustained virologic response, and where in that sequence does your population fall out. Funders backing an explicit doubling target need a baseline they can measure against. An organization that already knows its own numbers is a much lower-risk grantee than one that would have to build measurement from scratch.

Name the operational failure, not the technology. The strongest applications in this category identify a specific breakdown — positive antibody results never followed by RNA confirmation, patients lost between diagnosis in a correctional or emergency setting and treatment in the community, prior-authorization attrition — and then describe the tool that closes it. Applications that lead with the technology and search for a problem read as vendor pitches.

Show the data-sharing agreements you already hold. Identifying people overdue for treatment requires matching records across a health department, one or more delivery systems, and often a lab or a payer. The legal work behind that matching is slow. Executed agreements, or an active governance relationship with a state health department, are a stronger credential here than a technology plan.

Bring the partnership. The coalition's founding members are hospital associations, physician organizations, health plans and an integrated delivery system. This is an initiative that expects collaboratives, not solo applicants. A community organization with the trust of a hard-to-reach population and a health system with the clinical capacity are worth far more together than either alone.

Plan for what survives. The initiative's own theory of change is that pilots have to become permanent. Applications should say how the workflow persists once philanthropic money ends — billing pathways, health department budget lines, or system absorption. And expect the design to reach beyond hepatitis C, because HIV prevention and curable cancers are named as the next targets. Infrastructure that generalizes to another cascade is worth more than a single-disease tool.

The initiative also remains open to federal partnership, with leadership signaling willingness to collaborate as the work develops — a notable posture in a year when federal public health funding has been anything but predictable.

The larger pattern in this funder's portfolio

B2F is the third major disbursement in a short span from a philanthropy that has moved unusually fast. The OpenAI Foundation's $1 billion in grants established health and AI safety as its principal lanes, and its $50 million People-First AI Fund delivered unrestricted grants to community nonprofits on a compressed timeline. B2F is larger, narrower and considerably more demanding: it comes with a measurable two-year target attached to a specific clinical outcome.

For nonprofits and public health agencies, that is the trade to weigh. A hundred million dollars is real money aimed at a problem that is genuinely solvable, and the accountability is specific enough that grantees will be measured on cure rates rather than activities. Organizations that can already state their cascade numbers, hold their data agreements, and name their partners are the ones that will be ready when the next four jurisdictions are announced — and tools like Granted can help turn that operational knowledge into a proposal while the window is still open.