Regulators Draw Lines Around Health AI as Clinical Evidence Grows
WHO flags AI ethics gaps in global health research as regulators, trials, and drug developers push AI deeper into medicine.
By Dr. Asher Knippel
The week's health headlines converged on a single question: as artificial intelligence moves from research environments into hospitals, drug pipelines, and national screening programmes, do the institutions governing it still have the tools they need?
Tuesday, 17 September: FDA Maintains Clearance Requirement for Radiology AI
The US Food and Drug Administration published a final order declining to grant a partial 510(k) exemption for radiology AI software, reaffirming that manufacturers must continue to seek formal clearance before marketing these tools. The decision signals that the FDA does not yet consider AI-assisted imaging software a low-risk device category, even as the technology matures and hundreds of products have already cleared review.
On the same regulatory front, the Centres for Medicare and Medicaid Services (CMS) announced that Aidoc's CARE Body CT Multi-Triage AI tool will receive the first-ever New Technology Add-On Payment (NTAP) designation for an AI-enabled diagnostic, effective 1 October 2026. The three-year supplemental payment mechanism will provide hospitals with additional reimbursement when they deploy the tool—a landmark step toward sustainable financing for AI diagnostics in US hospital settings, and a signal to European payers and regulators about how such tools might eventually be priced.
Sunday, 21 September: WHO Calls for Stronger Ethics Oversight of AI Health Research
The World Health Organization released a new report warning that existing ethics infrastructure—institutional review boards, data governance frameworks, informed-consent procedures—was not designed for AI's particular risks and may be inadequate for the pace of current deployment. The report identifies algorithmic bias, health equity, privacy under large-scale data training, and the difficulty of applying traditional ethics review to iterative, continuously updated models as the central concerns.
WHO notes that most AI development remains concentrated in high-income countries, creating equity gaps for health systems in lower-income regions. The organisation recommends reforms spanning the full research lifecycle, with responsibilities distributed across researchers, ethics committees, regulators, and funders. For readers in Cyprus and the broader EU, where AI Act compliance timelines are tightening, the report offers a useful marker of where international consensus is heading. It does not constitute binding guidance, but its framing is expected to inform EU and national health-AI policy discussions over the coming year.
Sunday, 21 September: AbbVie and Iambic Partner on AI-Driven Small-Molecule Discovery
Pharmaceutical company AbbVie announced a multi-year collaboration with clinical-stage AI firm Iambic to apply Iambic's Enchant v3 platform—trained on measurements of more than 6,000 molecular properties—to the identification of first-in-class and best-in-class small-molecule drug candidates. The partnership targets oncology, immunology, and neuroscience simultaneously, aiming to compress the early-stage discovery process.
AI-assisted molecular design does not replace experimental chemistry or clinical trials; it narrows the hypothesis space by predicting which candidates are most likely to bind their targets and reach cells without unacceptable toxicity. The Enchant platform has been validated against known compounds; this partnership will test whether those laboratory benchmarks translate to programmes with genuine clinical ambitions. Results will not be apparent for several years, and no candidate has yet entered trials under this collaboration.
Saturday, 20 September: Tempus AI Acquires Personalis for Cancer Recurrence Monitoring
Tempus AI, a Chicago-based precision-medicine company, announced the acquisition of genomics firm Personalis. Where Tempus has historically focused on tumour profiling to guide treatment selection, Personalis's core technology is molecular residual disease (MRD) testing—the use of highly sensitive blood assays to detect circulating tumour DNA after treatment, providing an early signal of cancer recurrence before it becomes visible on imaging.
The combination moves Tempus toward a full-lifecycle oncology platform: diagnose, select therapy, monitor response, detect recurrence. MRD testing is an active area of clinical investigation; its utility—specifically, whether a positive test should prompt treatment changes and whether those changes improve patient outcomes—is still being established in prospective trials for most cancer types. Patients should discuss the appropriate role of liquid biopsy monitoring with their oncologist rather than seeking it independently.
Tuesday, 23 September: Oracle Launches Life Sciences Data Intelligence Platform
Oracle announced a cloud-native Life Sciences Data Intelligence platform combining real-world patient data, domain-specialised AI agents, and natural-language query interfaces, enabling pharmaceutical researchers to interrogate patient cohorts, run outcome analyses, and assist in clinical trial design without specialist data-engineering support.
The platform targets a persistent bottleneck in drug development: the gap between available clinical and claims data and the analytical capacity to use that data efficiently. Natural-language interfaces lower the barrier for clinical scientists who are not data engineers. Oracle is one of several large cloud providers competing aggressively for pharmaceutical research contracts; independent evaluation of platform accuracy and data-quality controls is not yet publicly available, and researchers should scrutinise any real-world evidence outputs with appropriate statistical care.
Across the Week: AI Mammography Trial and Early Drug-Safety Evidence
A September 2026 review published in the journal Cancers and reported by Forbes highlighted findings from a Swedish trial of over 105,000 women in which AI-assisted mammography reading raised cancer detection rates from 73.8% to 80.5% compared with standard double-reading. The trial ran within Sweden's well-characterised national screening programme and represents one of the largest prospective evaluations of AI as a second reader in breast cancer screening to date.
Separately, The Guardian reported on Isomorphic Labs—an Alphabet subsidiary—presenting evidence that AI-derived drug candidates show improved Phase 1 safety-trial success rates compared with conventionally discovered molecules. Isomorphic, which secured $2.1 billion in recent funding, plans first-in-human trials of AI-designed small molecules by the end of 2026. These are Phase 1 safety studies; evidence of clinical benefit remains years away, and the company's claims should be read in the context of substantial commercial incentives.
The mammography finding is the most immediately practice-relevant result of the week: it supports the case for AI as a second reader in national breast cancer screening programmes, a question that health regulators across the EU—including Cyprus—are actively evaluating. Pending formal guidance, AI-assisted reading is not yet standard of care in most jurisdictions.
This article is intended for journalistic and informational purposes only and does not constitute medical advice, diagnosis, or treatment recommendations. Readers should consult a qualified clinician before making any change to their health care or treatment plan.