AI in Health: Clinical Proof, Regulatory Lines, and the Drug Discovery Surge

UK Government and 115 Charities Sound the Alarm on AI Health Information

Britain's health secretary has warned the public not to rely on AI tools for medical diagnosis, a message amplified by an open letter from 115 charities calling for a formal safety framework governing health AI. The warnings arrived at the same moment the Medicines and Healthcare products Regulatory Agency (MHRA) announced an extension of its AI Airlock regulatory sandbox into live NHS environments. The dual pressure — civil-society concern from below and regulatory experimentation from above — reflects a wider tension in the UK about how quickly to embrace AI in clinical settings. The MHRA's approach, allowing monitored real-world testing before formal approval, is designed to accelerate safe deployment, but critics argue that voluntary sandboxes are insufficient without binding standards.

Evaxion's AI-Designed Cancer Vaccines Show Strong Immune Responses

Danish biotech Evaxion is presenting data at two major oncology conferences this month that together give the clearest clinical picture yet of its AI-Immunology platform. Phase 2 results for EVX-01, a personalised neoantigen vaccine for melanoma being shown at SITC 2026, demonstrate potent T-cell responses, suggesting the AI design approach can identify and encode the right tumour-specific targets. Separately, new data for EVX-04 — an off-the-shelf therapeutic vaccine for acute myeloid leukemia — will be presented at ASH in December. The pairing of a personalised and an off-the-shelf candidate, both driven by the same underlying AI platform, positions Evaxion as one of the more tangible examples of AI moving cancer vaccine development forward.

AI Cardiac Diagnostics Clear New Clinical Bars

Two studies published ahead of the American Heart Association's 2026 annual meeting set a new credibility floor for AI-driven cardiac care. Viz.ai's AI-ECG tool demonstrated an ability to flag hypertrophic cardiomyopathy — a serious and often undetected heart condition — in routine ECG reads that human clinicians had cleared as normal, raising the prospect of population-level screening. Meanwhile, a Lancet Digital Health analysis of cardiac MRI found that AI-derived measurements of cardiac structure and function marginally outperformed expert clinician measurements in predicting all-cause mortality. The word "marginally" matters: the difference is modest, but the direction of the finding — AI performing above expert human baseline — is a threshold the field has been working toward.

Roche, UCB, and UK Sovereign AI Fund Back AI Drug Discovery

Three deal announcements within 48 hours last week suggest that pharma's AI-drug-discovery phase is moving from pilot to production. Chai Discovery, whose Chai-3 protein structure model is a leading open tool in the field, announced a collaboration with Roche to integrate its AI into protein-therapeutic programs. UCB signed a partnership with Cradle to deploy generative AI across its biotherapeutics portfolio — an unusually broad mandate rather than a one-program deal. And Potential Sciences emerged from stealth backed by the UK Sovereign AI fund, claiming a 100× cost reduction in quantum-chemistry GPU calculations relevant to drug design. Taken together, these moves indicate that the AI-drug-discovery infrastructure layer is being absorbed into large-pharma operations rather than remaining a standalone startup sector.

Mayo Clinic and UC Berkeley–UCSF Deploy AI into Frontline Workflows

Two US health-system announcements this week marked the shift from AI-in-research to AI-in-care-delivery. Mayo Clinic Platform launched a new clinical decision-support tool covering 35 specialties, available to organisations participating in the platform network — representing one of the broadest specialty footprints for a single AI clinical tool in a major health system. At the same time, the UC Berkeley–UCSF Center for Edge Medicine received a $10 million Laude Moonshots grant to build open-source AI infrastructure integrating wearable and electronic health record data via JupyterHealth. The Centre is explicitly oriented toward frontline clinical settings, targeting the kinds of continuous patient monitoring that hospital informatics teams often lack the infrastructure to run.

Harvard Wins $7M for AI Cancer Co-Scientists; Insilico Opens Longevity AI

Harvard Medical School's AURORA project has received a $7 million grant to build agentic AI systems designed to work as scientific collaborators in cancer research — generating hypotheses, designing experiments, and iterating on results alongside human researchers. The project represents a move toward AI as an active participant in the discovery loop rather than a passive analysis tool. Separately, Insilico Medicine released Longevity Grok, an open-source personal health AI developed with HMS geroscientist David Sinclair that aggregates individual health records to personalise longevity and anti-aging recommendations. The two projects sit at different ends of the discovery-to-consumer pipeline, but both treat AI as something capable of doing genuine scientific and medical reasoning.

EU Awards €6.5M for AI Pandemic Vaccines; FDA Classifies Cervical Cancer Screening AI

Regulators and funders on both sides of the Atlantic took concrete forward steps this week. The European Commission's Horizon Europe programme awarded €6.5 million to a consortium including NEC OncoImmunity for the PREPAIR VAC project, which aims to use AI to anticipate viral evolution and design broad-spectrum vaccines capable of responding to future pandemic threats. In the US, the FDA issued a final rule formally reclassifying AI-powered digital cervical cytology imaging as a Class II medical device eligible for 510(k) clearance, reducing the regulatory burden for future cervical cancer screening AI products. Both moves represent governments creating structural conditions — funding infrastructure and regulatory pathways — for AI health tools to reach patients at scale.