Life Sciences CIOs Digest

AstraZeneca Is Rebuilding the Plumbing Before the AI

Its CIO calls a full SAP move the groundwork for AI. This week also showed how much of that plumbing now sits with someone else.

Life Sciences CIO Weekly • Coverage: November 9 to November 15, 2026


Not every important story this week was about AI itself. Several were about the groundwork underneath it.

  • AstraZeneca is moving its entire SAP estate to cloud ERP, and its CIO described the move as laying foundations for AI.
  • FDA asked industry how patient-level real-world data can reach the agency when sponsors often don’t own it. Comments are due January 11.
  • Siemens disclosed an authentication flaw in PCS 7, BATCH and Process Historian with no fix planned, days after France named an Iranian unit behind the cold-chain sabotage we covered last week.

As I read the risk stories, they lean the same way. More of that groundwork now sits with partners, from validated clouds to commercial data vendors, and that’s where much of this edition spends its time.

📌 A personal note from me at the end of this edition: an open invitation to join a live conversation with fellow life sciences IT leaders in early December. Jump to it →

🔍 The Quick Read

  • Lead: AstraZeneca is moving its whole SAP estate to cloud ERP and calls it groundwork for AI.
  • AI & Data: Lab Thread connects AI assistants to lab records, and DeepMind’s AlphaGenome 2 reads variants cell type by cell type.
  • Regulatory & Policy: FDA asked how real-world data can reach it, and WuXi left a Pentagon list ahead of BIOSECURE’s December 18 list.
  • Cybersecurity & Risk: Siemens won’t fix a PCS 7 flaw, France named Iran behind cold-chain sabotage, and new pharma leak claims surfaced.
  • Leadership & Operating Model: Merck KGaA’s HR agent became its AI template, and BMS merged research and development.

Reading something a colleague should see? Forward the whole edition or copy the section that fits their world. If this reached you from someone else, subscribe free here to get every edition, plus access to the back archive.


📌 This Week’s Lead

AstraZeneca moves its entire SAP estate to cloud ERP and calls it the foundation for AI ↗

A top-10 pharma is presenting its ERP modernization as groundwork for AI.

On November 10, SAP said AstraZeneca is moving from its legacy SAP ECC environments to SAP Cloud ERP Private through RISE with SAP (SAP News). The move covers AstraZeneca’s full SAP estate, one of the largest in life sciences, and sits inside its Axial Business Transformation Programme. The stated goals are to simplify and standardize its technology, harmonize data and connect processes across the company.

CIO Chris Taylor said SAP has been “a key part of how we run our operations for more than two decades,” and that this chapter builds foundations “to move faster and unlock greater value from AI.” The AI pieces, including SAP’s Joule assistant and agents in finance, supply chain and manufacturing, are areas AstraZeneca is “exploring.” SAP gave no contract value, user count or completion date (ERP Leaders). The same week, AstraZeneca cut about 600 US respiratory and immunology sales roles under its $1.5 billion savings program, Reuters reported (Reuters via AOL).

What to watch: If you’re making a similar case to your board, AstraZeneca’s order is clean core, harmonized data, then agents. Keep validated-system requalification and manufacturing cutover risk in the plan.


🤖 AI & Data

Lab Thread lets scientists’ AI assistants query lab records through an MCP server ↗

Approved AI assistants can now reach lab data directly, which makes the connector itself something to govern.

On November 10, UK-based Lab Thread launched a Model Context Protocol (MCP) server that connects assistants such as ChatGPT and Claude to its lab platform without exporting records or building custom integrations (Lab Thread). The company says data stays in its cloud until an authorized query, with tenant isolation, role-based permissions and audit logs it describes as aligned with 21 CFR Part 11. It named research teams at Mayo Clinic and UMass Chan Medical School among its users and opened a Boston office.

What to watch: Treat lab MCP connectors like any interface to research or GxP systems. Inventory and approve them, and check that an assistant inherits the user’s permissions and no more. Part 11 claims still need testing in qualification.

Google DeepMind’s AlphaGenome 2 predicts variant effects cell type by cell type ↗

Human-genetics teams get a sharper target-validation tool, and IT gets questions about what data leaves the firewall.

On November 11, DeepMind released AlphaGenome 2, which predicts how DNA variants affect gene regulation in specific cell types and states. It was trained on more than 107 million single-cell profiles across 68 tissues (Google DeepMind). On an immune-cell benchmark it found 30% of likely causal variants, against 16% for the original model. James Lee of the Francis Crick Institute told Nature it reproduced in about 30 minutes a disease mechanism that took his team “a year and a half of experiments” (Nature). Non-commercial use is open through the API. Commercial access is in private preview, and the model is for research use only.

What to watch: Commercial terms when the preview opens, including whether proprietary sequences or single-cell data used to condition the model stay inside your environment.

Pierre Fabre and Hansoh license AI discovery platforms, and Flagship merges two AI biotechs ↗

The pattern holds: the vendor’s platform designs the molecules, and the sponsor keeps the program.

On November 9, Pierre Fabre and XtalPi announced a multi-program deal. XtalPi’s AI and robotics platform will design small molecules against hard-to-drug oncology targets, and Pierre Fabre will take compounds into clinical development. Terms were not disclosed (PR Newswire). On November 13, Hansoh Pharmaceutical said it licensed AilsynBio’s AI platform for oral cyclic peptides. Hansoh selects the targets and acquires the resulting IP, for payments of up to RMB 317 million plus royalties (HKEX filing). And on November 11, Flagship Pioneering said it is merging Valo Health and Montai Therapeutics under the Valo name (World Pharma AI).

What to watch: When AI vendors merge, check change-of-control clauses in partner data and model agreements, and how design data flows back into your own registration and ELN systems.

Anthropic sets IPO terms and calls drug discovery one of its largest efforts ↗

A core AI supplier to pharma is also building its own drug pipeline.

On November 10, Anthropic set an IPO price range of $70 to $78 a share, with trading expected November 19 (CNBC). The next day its head of life sciences, Eric Kauderer-Abrams, told Forbes the life sciences team is one of the largest efforts in the company (Forbes). Forbes reported that Anthropic bought Coefficient Bio for $400 million, opened a wet lab in San Francisco and is building a preclinical pipeline in rare, tropical and neglected diseases, with plans to partner with pharma for clinical work.

What to watch: Vendor risk reviews for AI model providers can now cover training on customer data, confidentiality of discovery work and conflicts of interest.


⚖ Regulatory & Policy

FDA asks how patient-level real-world data can reach the agency when sponsors don’t own it ↗

The data behind real-world evidence is often licensed, and FDA is asking industry how to get it to reviewers.

FDA published a request for information on November 12 about submitting the patient-level real-world data behind evidence in drug and biologic applications (docket FDA-2026-N-10742). Comments are due January 11, 2027 (Federal Register).

The notice is candid. Sponsors often license this data and may not have the right to pass patient-level records on. FDA often can’t sign vendor data-use agreements, because their destruction, ownership and sharing terms clash with federal records rules and FOIA. It also doesn’t accept links to outside databases. Today’s workarounds are a third party’s pre-IND or a Type V drug master file. FDA flags the size of imaging and multi-omics datasets too, and invites EDC and health IT vendors to respond (RAPS).

What to watch: Check whether your real-world data licenses allow regulator access and what their destruction clauses require. Trade associations are the natural channel for comments.

Pentagon removes WuXi AppTec from its Chinese military companies list, with BIOSECURE’s first list due December 18 ↗

One path to a BIOSECURE designation closed this week. Another comes due in five weeks.

The Defense Department published WuXi AppTec’s removal from the Section 1260H list on November 10. It cited an August court injunction and kept the right to relist the company “on additional or supplementary grounds” (Federal Register). Under the BIOSECURE Act, the Office of Management and Budget can still name WuXi as a biotechnology company of concern through a separate pathway. Its first list is due December 18 (Ropes & Gray).

What to watch: Keep vendor-risk records, exit clauses and data-transfer maps for Chinese CDMOs and CROs ready to review on short notice.

FDA warns three Chinese API makers, two for refusing electronic records requests ↗

A refused remote records request now moves quickly to an import alert.

FDA posted warning letters on November 10 to three Chinese API makers whose products are now under import alert (RAPS). Tianjin Healthcare Medical Technology and Chongqing Qiancan Technology declined electronic records requests sent in June. One said it would stop serving the US market, and FDA replied that stopping distribution “does not retroactively relieve you of your regulatory obligations.” The third firm, Nanjing Hanxin, complied but was cited for failing to qualify a critical raw-material supplier.

What to watch: Check that supplier qualification files and audit trails, including those held by your suppliers’ suppliers, can answer an FDA records request within days.

EU pharmaceutical package clears Parliament’s health committee ↗

A 24-month transition starts once the law is published, but regulatory sandboxes apply right away.

On November 10, the European Parliament’s public health committee adopted its second-reading recommendations on the pharma package, a regulation and a directive. A plenary vote still has to follow (EU Perspectives). Regulatory data protection starts at eight years and can reach 11 under conditions that include running clinical trials in Europe. Baseline orphan exclusivity drops from 10 years to nine. After publication in the Official Journal, a 24-month transition begins, while regulatory sandboxes apply on entry into force.

What to watch: Regulatory information, launch tracking and trial-location data may all feed the new protection conditions. Map those dependencies before the clock starts.


🔒 Cybersecurity & Risk

(Unverified) Ransomware group claims 3 TB from a validated life sciences cloud ↗

If the claim is true, one intrusion could reach regulated documents for many sponsors at once.

On November 10, the tracker Ransomware.live recorded a Black X leak-site listing for Court Square Group. The group claims “3T of customer data from the RegDocs365 and ARCC cloud network,” including clinical trial data and research materials (Ransomware.live). Court Square sells RegDocs365 as a “fully 21 CFR Part 11 compliant” document platform, hosted in a validated cloud for pharma, biotech and medtech companies and CROs, including multi-sponsor setups (Court Square Group). We found no company statement, and the claim is unverified.

What to watch: Find out whether any business unit, CRO or partner uses RegDocs365 or ARCC, ask for a written incident statement, and check breach-notice terms in your validated-cloud contracts.

Siemens says no fix is planned for an authentication flaw in PCS 7, BATCH and Process Historian ↗

Core plant systems that hold GMP batch data will rely on a workaround.

On November 10, Siemens published an advisory for CVE-2026-58114 (CVSS 7.5). The weak-authentication flaw lets an attacker on the network path hijack sessions in SIMATIC PCS 7 V9.1, SIMATIC BATCH V9.1, Process Historian 2020 and 2022 and several WinCC versions (Siemens). No fix is planned for PCS 7, BATCH or Process Historian. The workaround is to turn on encryption in the underlying SQL Server and restrict network access. Siemens also flagged Node2Shell, a 9.8-rated flaw in a decoder bundled with Node.js, in WinCC and other products (SecurityWeek).

What to watch: Apply the SQL Server workaround under change control, record it in the validated-state file, and confirm segmentation between plant and enterprise networks.

France names an Iranian IRGC unit behind the refrigeration controller sabotage ↗

The cold-chain attacks we covered last week now have a nation-state name attached.

On November 9, France’s cybersecurity agency ANSSI and its foreign ministry attributed the October attacks on refrigeration controllers at about 700 French retail sites to Shahid Kaveh, a unit of Iran’s Islamic Revolutionary Guard Corps. The Paris prosecutor requested arrest warrants for three commanders (The Record). Attackers changed defrost settings, causing stock losses and at least 15 fires. The early-November tampering at three French blood service sites has not been attributed to the same group, and Iran denies involvement.

What to watch: Freezer farms, biorepositories, stability chambers and 3PL cold chains use the same class of controllers. Ask facilities and logistics partners to confirm none can be reached from the internet.

(Unverified) Extortion group names Vertex and LEO Pharma, claiming commercial analytics data ↗

The claims target prescriber and patient-support data, which often sits with vendors outside core security monitoring.

Leak-site tracker Ransomware.live recorded the N0n group naming Vertex Pharmaceuticals on November 9 and LEO Pharma on November 12 (Ransomware.live). N0n claims a Vertex prescriber list with more than 248,000 named nurse practitioners and physician assistants and 237,000 DEA numbers. For LEO it claims a “patient companion analytics corpus” of 315 million records plus 441,000 patient-support transactions (DeXpose). Neither company has confirmed an intrusion, and the counts are the attackers’ own.

What to watch: Inventory where prescriber files, patient-support records and licensed claims data live, including at hub and analytics vendors, and confirm who owns breach notice.

Dexcom confirms a vishing attack exposed employee and patient files ↗

Phone-based social engineering reached document stores holding identity and health data.

On November 12, Dexcom said targeted vishing (voice phishing) attacks led to unauthorized access to internal documents that were later published on an illicit website. It said the incident is contained and its products are unaffected (Dexcom). Its California attorney general filing lists a November 2 breach date and two notices. One covers Social Security numbers and government ID, and the other covers health information such as prescriptions and certain diagnoses (California AG). Dexcom did not say how many people were affected.

What to watch: This incident tests phishing-resistant MFA, call-back checks on help-desk resets, and alerts on large file downloads.

NIST draft adds an agentic AI attack taxonomy, with comments due December 18 ↗

Teams putting agents into GxP-adjacent work now have a shared vocabulary for threat models.

On November 9, NIST released a draft update of its adversarial machine learning guide, NIST SP 2400-102e2026. NIST says the draft focuses on attacks on agentic AI systems and their mitigations, and it adds a new index of agentic attacks alongside the main taxonomy (NIST). Comments close December 18 (NIST CSRC).

What to watch: Use the draft’s terms to frame threat models and validation tests for agents in batch record review, regulatory drafting or safety intake.


🎯 Leadership & Operating Model

Merck KGaA made its HR agent the template for enterprise AI, and identity was the critical path ↗

The function with the most mature data went first, and solving access rights took months.

In a November 12 interview with UNLEASH, Alexis Saussinan, Group Head of People Insights & Effectiveness at Merck KGaA, Darmstadt, described how V.E.R.A., its agentic HR assistant, became the company’s model for AI (UNLEASH). Every AI initiative must now pass four tests: a clear business problem with measurable value, end-to-end workflow redesign, enterprise readiness, and people at the center. Employee authentication and access rights “took a few months to crack.” Savings came mainly from consultant and IT tool spending, not headcount. The open question now is how to “look for, hire, interview, onboard, contract and retire an agent.”

What to watch: Managing an agent from approval to retirement is likely to land on IT, HR and risk together.

Bristol Myers Squibb merges research and development into one organization ↗

Separately funded discovery and clinical platforms will now share one R&D roadmap.

On November 12, BMS said it will combine research and development into a single organization on January 1, 2027. It will be led by Cristian Massacesi, now chief medical officer and head of development (BMS). Chief research officer Robert Plenge will advise through March and then leave. CEO Christopher Boerner said the structure should help BMS attract talent, build capabilities and integrate new technology (BioPharma Dive).

What to watch: Expect pressure to connect preclinical and clinical data end to end, and watch whether BMS names one R&D data or digital leader.

Moderna puts AI and automation behind a flat $4 billion cash-cost plan ↗

A biopharma is tying a multiyear cost envelope to AI in business operations as well as discovery.

At its November 12 analyst day, Moderna said it expects cash costs of about $4 billion in 2027 and $4 billion to $4.5 billion a year through 2030, and raised its 2026 revenue growth outlook to 15% to 25% (Moderna). CEO Stéphane Bancel said “artificial intelligence and automation are accelerating what our platform can achieve,” and the company said both are being deployed across manufacturing and business operations (Pharmaceutical Executive).

What to watch: Peers will be measured against this story. Watch for operating metrics that back it up.


🤝 A seat at the table: join a live conversation in early December

Reading about these issues is one thing. Talking them through with other IT leaders who are dealing with the same things is something else entirely.

A few times a year, I bring together a small group of life sciences IT leaders to talk through the signals that keep showing up in this digest. We’ve done this several times before and had some great conversations. These sessions usually happen inside my community and aren’t open, so this is a chance to join in with peers who have been there, and whose background and experience you can trust.

Since late summer these pages have covered proving what the AI did, a mandate that keeps growing faster than the budget, less time to respond when something breaks, and AI partnerships that turn out to be data deals. I’d like to talk through what those signals mean for your 2027 plans, and hear what you’re seeing that the headlines miss.

The plan is about an hour, online, in early December. It’s for peers only, with no vendors and no pitch.

If you’d like a seat, just click one of the options below. I’ll follow up personally with the details.

Would you like a seat at the live session in early December?

One click is all it takes. Joe will follow up personally with details.

Login or Subscribe to participate

Joe

Ready to move beyond the digest? The LS CIO Community is where these conversations continue.

Join the LS CIO Community →


Subscriber sources (often licensed inside life sciences companies): Bloomberg, WuXi AppTec removed from Pentagon list (Nov 10); Pink Sheet, WuXi AppTec removed from 1260H list; Bloomberg, AstraZeneca US sales cuts (Nov 13); Endpoints, BMS merges R&D (Nov 12).

How this is made: each edition is researched two ways in parallel, once with Perplexity and once with Claude, then reconciled into a single verified brief.

This digest is an interpretive summary of publicly available information and does not constitute legal, regulatory, cybersecurity, or investment advice.

Until next week,

Joe Miller

Founder, Leadership Inklings