Genentech Signs $1.5B+ Bispecifics Deal With AI Biotech Earendil Labs 🔗

On 24 September, Earendil Labs announced a research collaboration with Genentech to discover and develop multiple bispecific antibody programmes in oncology. Earendil leads antibody discovery through early clinical development; Genentech takes global clinical development and commercialisation. The deal pays $55 million upfront, with total potential value, including development, regulatory and sales milestones, in excess of $1.5 billion, plus tiered royalties. Earendil's platform combines predictive protein modelling, generative protein design and rapid experimental validation. Within the same week, Earendil was named in Fierce Biotech's AI heavyweight coverage alongside Iambic and Xaira, signalling that generative biologics discovery platforms are now competing for the biggest biopharma dollars.

AbbVie Signs Iambic for AI-Driven Small Molecules; Iambic Files for IPO Days Later 🔗

On 21 September, AbbVie and Iambic announced a multi-year collaboration applying Iambic's molecular superintelligence platform to small molecule discovery in immunology, neuroscience and oncology. Iambic will use its newly launched Enchant v3 platform under the deal. The following day, Iambic filed for an IPO on Nasdaq, aiming to fund its lead candidate IAM1363 (an oral HER2 inhibitor in Phase 1 for HER2-positive breast cancer and NSCLC) plus two preclinical programmes. The combined move puts a reference point under the category: a clinical-stage AI drug discovery company with a big pharma deal in hand going public to fund its own pipeline, rather than remaining a platform-for-hire.

The VA Awards $775.72M Ambient Scribe IDIQ to Abridge and Knowtex 🔗

On 18 September, the VA Technology Acquisition Center awarded a five-year, multi-award Indefinite Delivery Indefinite Quantity contract for ambient scribe AI across the Veterans Health Administration, with a shared ceiling of $775.72 million. Abridge AI and Knowtex were named the two prime awardees. Both companies had been running under VA pilots since October 2025 and are now in production at over 75 VA medical centres. Veterans Health Administration already reports that over 986,000 primary care appointments have used ambient scribes since those pilots started, with veterans declining the technology less than 1% of the time. This is the largest public enterprise commitment to ambient AI in American healthcare to date, and it sets a procurement pattern (multi-award IDIQ, VISN-level task orders, strict KPIs on uptime and support) that other federal and state health systems will likely copy.

Federated AI Beats Public Co-Folding Models for the First Time at Scale 🔗
The AI Structural Biology Network, powered by Apheris with the AlQuraishi Lab at Columbia, published results from its Federated OpenFold3 Initiative. In under ten weeks, AbbVie, Astex, Bristol Myers Squibb, Johnson & Johnson and Takeda fine-tuned OpenFold3 inside their own environments on over 20,000 proprietary protein-ligand structures, roughly three times the drug-relevant data in any public model. The resulting AISB-1-Fed outperformed public Boltz-2 and every model fine-tuned on a single partner's data. Four of the five have signed on to a follow-up initiative, AISB Bind, starting this month.

The signal: This is what the Lilly TuneLab story in the last issue was pointing at, now with a published performance number. Pharma's proprietary structural data has sat in silos for twenty years because nobody could share it, and is finally being made useful without being moved. The business model to watch is Apheris itself: whoever operates the federation layer sits between every participating pharma and every next model.

Veeva Puts Agents into Clinical Trial Setup and Deepens the Vault CRM Lock-In 🔗

Veeva confirmed a 14th top-20 biopharma commitment to Vault CRM, which now has over 190 live customers. It also announced Study Builder Agent, which configures Veeva EDC and the Data Quality System directly from a study protocol using the CDISC Unified Study Definitions Model, delivered as a Claude Cowork plugin from a Veeva-managed GitHub repository, with early access in December. Servier separately adopted Veeva OpenData globally across 80+ countries to standardise customer reference data as the base layer for commercial AI.

The signal: Veeva's thesis is that AI usefulness in regulated life sciences is bounded by data quality and system standardisation first, and the company is methodically putting agents inside Vault, EDC, DQS and OpenData. Study Builder Agent is also the first mass-market example of the Anthropic Model Hardware Standard pattern in production: an agent packaged as a Cowork plugin, versioned in a vendor-controlled Git repo, running inside the application it configures. Expect more life sciences software vendors to ship AI features in this shape.

ARPA-H Keeps Spending: $38M for ICU Digital Twins, $35M for Rare Disease AI 🔗

The University of Vermont received up to $38 million for ReSCUED, a project to build real-time digital twin AI for ICU patients, with validation planned over three years. A consortium led by UNC School of Medicine and Emory University, with Global Genes joining, received up to $35 million under ARPA-H's RAPID programme (Rare Disease AI/ML for Precision Integrated Diagnostics), anchored on Global Genes' RARE-X patient-powered registry of 10,000+ participants.

The signal: ARPA-H is running a consistent playbook across autonomous clinical AI (cardiology last issue, now ICU) and rare disease AI. Each programme pairs federal capital with a mandated validation study, and in several cases a planned FDA pathway. AI health companies that want in on the next programmes will need to build validation designs around what ARPA-H is funding, not around their own roadmap.