Collate Raises $95 Million to Automate Regulatory Documentation 🔗

Collate, founded 18 months ago, has raised $95 million in a Redpoint-led round and is approaching a $1 billion valuation. The company's AI reduces document collation time for regulatory submissions by 50 to 90%, with human oversight built into every output before anything reaches a regulatory body. Around 50 enterprise pharma and biotech clients are on the platform. The category — AI-assisted regulatory documentation — has now produced a near-unicorn in under two years.

Microsoft and Mayo Clinic Announce Frontier Healthcare AI Model 🔗

Microsoft and Mayo Clinic are jointly developing a frontier AI model trained on Mayo's longitudinal clinical data, designed for diagnostic accuracy and treatment decision support. Ownership remains with Mayo Clinic. Microsoft plans to make the model available via Azure Foundry APIs to other healthcare organizations. The structure — clinical institution as owner, tech company as infrastructure provider — reflects the data sovereignty tension now defining every major AI-in-healthcare partnership.

Agilent Partners with OpenAI and BCG to Scale AI Across Its Business 🔗

Agilent Technologies is deploying AI across its instruments, customer workflows, and internal operations through a partnership combining OpenAI's model capabilities and BCG's enterprise transformation expertise. The initiative targets smarter analytical instruments and optimized internal processes at scale. For a company whose core business is laboratory hardware and consumables, this is a signal that the intelligence layer is becoming inseparable from the physical instrument.

FDA Is Now Citing AI Explainability in Deficiency Letters — Not Just Performance 🔗
The FDA is issuing deficiency letters to AI/ML medical device manufacturers not for underperforming models, but for failing to explain how their algorithms reach their outputs. The shift is documented: manufacturers optimizing for sensitivity and specificity metrics are now receiving requests for traceability of model logic, subgroup performance analysis, and evidence that clinicians can meaningfully interpret and override AI outputs. The FDA's existing guidance documents collectively establish a transparency framework that predates a specific "explainability guidance" label — but the enforcement signals are now arriving.

The signal: This is the governance story that connects the full week. Collate built human oversight into its core architecture. Owkin's K Pro is being deployed because it generates traceable scientific reasoning. The FDA is making explainability a compliance requirement. The companies that built accountability into their AI stack from the start are structurally ahead.

AI Governance in Biotech CMC Is Fragmented — and the Regulator Is Watching 🔗

A Bioprocess Online analysis this week documented a consistent pattern across biotech: AI is being used actively in Chemistry, Manufacturing, and Controls and regulatory affairs workflows, but governance frameworks are patchwork — particularly at smaller companies under pipeline pressure. Different AI tools are being used without centralized oversight, creating traceability gaps in AI-generated regulatory outputs. Larger pharma has moved further on governance frameworks; biotech is largely in pilot phase without the controls to match.

The signal: This is the shadow AI problem that Sapio Sciences flagged last month in ELN workflows, appearing now in manufacturing and regulatory contexts. The difference: CMC outputs go directly into regulatory submissions. A traceability gap here is not a lab notebook audit risk — it is an inspection risk and a submission risk. Vendors who solve CMC governance at the infrastructure level, not as a bolt-on, are entering a market where the pain is real and documented.

Optellum Passes 250 Clinical Sites and 3 Million Cases in AI Lung Cancer Diagnostics 🔗

Optellum's Lung Cancer Prediction AI has now been deployed at over 250 clinical sites and processed more than 3 million cases, including partnerships with Bristol Myers Squibb and the NHS. The company's Virtual Nodule Clinic assists a patient every 14 seconds. FDA clearance and Medicare reimbursement are in place. Its LungOS platform includes the world's first Thorax CT Foundation Model. LungDetect, targeting Europe, is in launch preparation.

The signal: Three million cases is not a pilot number. Optellum is past the proof-of-concept phase that still occupies most of the AI diagnostics market — it is in the deployment phase that determines clinical standard of care. The combination of Medicare reimbursement, NHS scale, and pharma partnerships gives it the data loop that smaller competitors cannot replicate. Leica Biosystems expanded its AstraZeneca and Daiichi Sankyo collaboration on AI-driven lung cancer biomarker research this same week, using digital pathology and image analysis algorithms for TROP2. The convergence of imaging AI and biomarker research in oncology is moving from parallel tracks to integrated infrastructure.