Dear friends,
E.B. White once described New York as several different cities occupying the same geography. During United Nations General Assembly High-Level Week, that feels particularly apt.
There is the New York of diplomacy – presidents, ministers, ambassadors and carefully negotiated language. There is the New York of science, technology, finance and industry, where ideas and innovation often move considerably faster. And then there is the New York that matters most for health: the New York of implementation – patients, clinicians and health systems asking what any of this will actually change once everybody goes home.
For a few days, all these New Yorks share the same streets. And everyone arrives carrying an idea.
Perhaps Inception provides the modern twist. Ideas about AI, cancer screening, data, regulation, financing and access are being seeded everywhere. Some will prove transformative. Some may be premature. Others work wonderfully in one system and rather less wonderfully when they encounter another.
Health diplomacy is partly the art of navigating those ideas – testing ambition against evidence, culture, affordability, patient benefit and the realities of implementation. And this is where IAPM sees an important role: bridging the people generating the science with the people who must make it work, and bringing ministers and ambassadors into the same conversation as patients, clinicians, researchers, regulators, health-system leaders and innovators.
That thread runs through this edition of the IPM Brief: from what “early” really means in cancer detection and AI entering the wet lab, to medicines designed around one patient, biological tipping points in Alzheimer’s and the increasingly blurred frontier between AI co-scientists and AI co-policy makers.
Independent media have a role in testing these propositions too. CNN, MS NOW and Politico have gone to court after being excluded from the White House, arguing that the decision violates First Amendment protections; the administration has defended its actions. (Reuters) Whatever the eventual legal outcome, independent scrutiny matters. So should any correspondents suddenly discover a little more flexibility in their diaries this week, there will certainly be room for an independent pair of eyes at our New York discussions.
First, of course, everybody has to get there. A telecommunications failure compounded by a severed fibre-optic line disrupted thousands of flights across the United States, particularly in the Northeast, before operations resumed. (Reuters)
For now, the IAPM flights remain on schedule. Which is fortunate. Because, as the agenda puts it: Personalisation without access is simply better science describing the same inequality.
Many New Yorks. Many ideas. And two great IAPM events. The question behind all of them: Which ideas can survive the journey from ambition to reality?
| NEW YORK, THIS WEEK 23 September, 14:00-18:15 | MFN, Trade and Tariffs: The Future of Personalised Medicine | 777 United Nations Plaza |
| 24 September, 09:00-13:00 | Bringing Innovation into Healthcare Systems | Italian Cultural Institute, 686 Park Avenue Register for one or both meetings |
In this issue
- GRAIL’s Galleri reaches the FDA — and the word “early” becomes the battleground: A blood test that can detect many cancers now faces the harder question of which cancers it finds before they spread.
- When AI enters the wet lab, who gets to do the science? Anthropic is building physical laboratory capacity while verification rules begin deciding who may use advanced biological AI.
- Don’t cut the DNA. Switch it off. An experimental hepatitis B therapy is trying to silence persistent viral DNA without changing its sequence.
- One patient. One genetic variant. One medicine. A personalised ALS therapy turned one mutation into a one-person development programme – and a regulatory test case.
- Alzheimer’s may be a disease of tipping points: A high-resolution brain map suggests that timing and biological state may matter as much as diagnosis.
- From AI co-scientists to AI co-policy makers? If machines can challenge scientific hypotheses, they may soon help governments stress-test policy options too.
- Keep AI under human control — but do we really need another institution? Global leaders want stronger frontier-AI safeguards; the institutional architecture still needs a diagnosis.
GRAIL’s Galleri reaches the FDA — and the word “early” becomes the battleground
The FDA’s Molecular and Clinical Genetics Panel will examine GRAIL’s Galleri test on 23 September, moving the multi-cancer early-detection debate from promise into label language. In PATHFINDER 2, Galleri showed 35.0% overall sensitivity, 99.85% specificity and 77.0% positive predictive value. In NHS-Galleri, overall sensitivity was 31.6%; it rose from 13.6% in stage I to 60.0% in stage IV, while stage I-II sensitivity was 20.5%. GRAIL has also acknowledged that the NHS trial did not meet its primary endpoint of reducing the combined incidence of stage III and IV cancers, although it reported encouraging later-round and stage IV signals. (US Food and Drug Administration; FDA advisory meeting; GRAIL)
The distinction is the entire regulatory problem. A test may detect signals associated with more than 50 cancers without detecting every cancer equally well, or equally early. That does not make Galleri a failure; it makes the word “early” a claim that must be earned cancer by cancer, stage by stage and population by population. Multi-cancer screening could change oncology. Its credibility will depend on a label that tells patients and health systems what the evidence actually shows, not what the category name invites them to imagine. The next generation of cancer screening may begin with a blood test. The next generation of regulation begins with getting the noun and the adjective right.
When AI enters the wet lab, who gets to do the science?
Anthropic has established a wet laboratory in the San Francisco Bay Area as it expands beyond computational biology towards physical experimentation and automation. It has not published a full equipment inventory, but the direction is clear: AI systems will increasingly move from reading papers to proposing experiments, interacting with laboratory workflows, analysing results and designing the next round. At the same time, Anthropic has launched a Life Sciences Verification Program that checks credentials, security standards and ethical oversight before granting some users access to more capable biological functions. (Reuters; Anthropic)
There are legitimate safety reasons for this. Advanced biological AI is inherently dual-use. But the wet lab turns the governance question into a question of ownership: who controls the instruments, the training data and the permission to participate? If frontier biology increasingly depends on proprietary models, proprietary experimental systems and proprietary datasets, open science cannot be measured only by who may read the final paper. It must also ask who was allowed to produce the discovery. Safety should reduce dangerous access. It should not quietly become a velvet rope around the scientific frontier.
Don’t cut the DNA. Switch it off.
CRMA-1001 is an experimental hepatitis B therapy designed to silence persistent viral DNA without cutting the DNA sequence itself. It uses a catalytically inactive Cas9 – CRISPR without the molecular scissors – to guide methylation machinery towards both cccDNA and integrated HBV DNA. Company-reported preclinical studies showed deep, durable reductions in viral biomarkers, and a first-in-human Phase 1/2 trial is now evaluating safety, tolerability and early biological activity. The target is enormous: WHO estimates that around 240 million people were living with hepatitis B in 2024 and that the infection caused approximately 1.1 million deaths. (nChroma Bio; ClinicalTrials.gov; World Health Organization)
Animal success is not human proof, and “durable” in a mouse is not “functional cure” in a clinic. But the conceptual shift matters. Traditional genome editing asks whether we can change the sequence. Epigenetic medicine asks whether we can leave the sequence intact and change what it does. If that control proves precise, lasting and safe, the implications extend far beyond hepatitis B. Medicine would gain a new political verb for DNA: not delete, not replace, but silence.
One patient. One genetic variant. One medicine.
Researchers have reported early results from nL-CHCHD-001, an antisense therapy designed around a pathogenic CHCHD10 variant carried by a single patient with ALS. The programme screened 320 gapmer antisense oligonucleotides before selecting the candidate. After six intrathecal doses over 12 months, plasma neurofilament light fell by about 50% into the normal range, the ALS Functional Rating Scale-Revised moved from 33 to 36 and predicted vital capacity rose from 48% to 55%. No serious adverse events were reported. It remains one patient, without a control group, and cannot establish efficacy. (Brain Medicine; PubMed; US Food and Drug Administration)
The larger experiment is the development model. A molecular diagnosis became a bespoke medicine in roughly three years, while the FDA’s draft Plausible Mechanism Framework is beginning to describe how individualized therapies might be evaluated when conventional randomised trials are impossible. Traditional development asks how many patients can be recruited for one drug. Individualized therapeutics ask whether a drug can be built around the molecular cause in one patient, and whether evidence can be constructed honestly around that reality. The medicine is personalised. Regulation, manufacturing, reimbursement and liability will now have to become personalised enough to meet it.
Alzheimer’s may be a disease of tipping points
A multimodal map of the human hippocampus has added anatomical weight to the idea that Alzheimer’s disease progresses through biological states rather than one smooth cascade. Using postmortem tissue from octogenarians and centenarians, researchers identified six tissue domains and two major inflection points, including a transition from amyloid-associated inflammation towards phosphorylated-tau-associated degeneration and microglial change. Different regions may cross these thresholds at different times, producing a mosaic of disease rather than one uniform brain state. (Nature Medicine; Neuron)
This is not yet a clinical staging tool, and postmortem maps cannot tell us exactly how an individual patient will move through disease. But the policy implication is already uncomfortable: a diagnosis may be too coarse for a biology that changes state. The question shifts from “Does this patient have Alzheimer’s?” to “Which transition is this patient approaching, and what intervention is still capable of changing the trajectory?” Blood biomarkers, imaging, genetics and molecular signatures may eventually locate that state. In Alzheimer’s, timing may not merely influence treatment. Timing may become part of the treatment itself.
From AI co-scientists to AI co-policy makers?
AI co-scientists are moving beyond literature search and summarisation. Multi-agent systems can generate hypotheses, challenge competing explanations, rank proposals and iteratively refine experiments; a peer-reviewed evaluation of Google’s AI co-scientist has now moved that idea from demonstration towards scientific scrutiny. In parallel, the United Nations has established an Independent International Scientific Panel on AI and a Global Dialogue on AI Governance to provide governments with a shared evidence base and an international forum. (Nature; Google Research; United Nations)
Now imagine giving the same architecture a policy problem: how do we increase cancer-screening uptake without widening inequality? Different agents could interrogate epidemiology, budgets, reimbursement, workforce, behaviour and law; construct rival options; expose assumptions; and model unintended consequences. The human policymaker would still decide. The useful model is not a machine replacing political judgement, but a political system forced to inspect more evidence before exercising it. Governments have spent years asking how to regulate AI. They may soon have to ask how to use it without letting analysis become authority.
Keep AI under human control — but do we really need another institution?
A call led by Finland and Norway, supported by leaders from 20 countries, argues that frontier AI must remain under human direction, oversight and control. It calls for transparent safety protocols, mandatory pre-deployment testing, independent evaluation, incident reporting and stronger international coordination, while asking states to explore institutional arrangements for standards and verification. The principle is sensible. The architecture is where the politics begins. (President of the Republic of Finland; Government of Norway; United Nations)
The international landscape is not empty. The UN already has a scientific panel and a global dialogue. Before creating another body, governments should identify what those mechanisms cannot do and build precisely for that gap. Health adds another warning: data come with language, culture, law, clinical practice and different tolerances of risk. We need global rules where genuinely global risks demand them, but interoperability must not become uniformity. Finland does, of course, have Santa Claus in Lapland, so perhaps there is room for a little Christmas magic. No institution, however, can wave away local reality. The task is not more governance. It is governance strong enough to manage transnational risk, flexible enough to respect context and coordinated enough not to become another waiting room.
Back to New York
The question is not whether innovation will continue. It will. The question is whether our regulatory systems, healthcare systems and international institutions can distinguish what is genuinely transformative from what is merely technologically impressive – and then translate the former into patient benefit.
Many New Yorks. Many ideas. Two IAPM meetings. And one recurring challenge: How do we get from what is scientifically possible to what is practically available?
Perhaps the best way to answer that is not from one perspective, one sector or one institution. Join us at the IAPM events in New York. IAPM brings together ministers, ambassadors and policymakers who can shape the environment; patients and clinicians who live with its consequences; and the researchers, innovators and movers and shakers of the personalised-medicine world who are pushing the science forward.
Because bridging science and policy is only the beginning. The real test is whether we can bridge both to implementation.
Registrations also remain open for upcoming IPM Alliance events in New York on 24 September and Dublin on 2 October. Visit the IPM Alliance website for programme updates.
Follow the last-mile conversation!
IPM Alliance tracks the decisions, bottlenecks and implementation gaps shaping personalised medicine access. Follow us for briefings, events, policy signals and practical insights from across regions and disease areas.
We want to hear from you
When a patient becomes eligible for personalised care, where does the system most often lose them: identification, testing, referral, reimbursement, delivery capacity or equity?
Know someone who should be reading this?
IPM Brief is built for people working where science, policy and patient access collide. If one colleague would find this useful, forward it to them and help bring them into the conversation.
International Alliance for Personalised Medicine
Avenue de l’Armée / Legerlaan 10, 1041 Brussels, Belgium
