AI Moves Into the Real Lab

AI is starting to run real experiments, not just simulations. That speeds the hunt for drug candidates, and it runs on your lab data.

Life Sciences CIO Weekly • Coverage: October 5 to October 11, 2026

AI in drug discovery stepped off the screen this week and into real experiments. Danaher announced a lab where AI picks the tests and robots run them. AstraZeneca is building the same kind of loop with Carterra, and GSK tested Chai Discovery's AI in its own labs before it signed. The goal is speed: find good drug candidates sooner and beat rivals to market. Regulators kept pace, with the UK and the FDA both setting up new structures for AI. On the risk side, U.S. agencies exposed a Chinese hacking contractor, and an unpatched flaw put in-house AI servers at risk.

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 and, coming soon, our CIO acronym glossary.

🤖 AI & Data

AI moves out of the simulator and into the lab

Three moves this week put AI in charge of real experiments, and all three depend on clean, connected lab data.

For years, most AI in drug discovery worked on a screen. Models predicted which molecules might work, and scientists tested a handful by hand. This week the loop closed. On October 7, Danaher announced an autonomous lab at its Abcam unit. AI picks the experiments, robots run them, and the results retrain the model. It draws on Danaher companies Beckman Coulter, Cytiva, Genedata, IDT and Molecular Devices, with robotics from Automata. Danaher projects up to eight times faster progress from idea to tested reagent and expects full capacity in early 2027 (Yahoo Finance).

On October 8, AstraZeneca and Carterra, a maker of high-speed instruments that measure how proteins bind, agreed to co-build lab hardware and software for the same kind of loop (News-Medical). On October 9, GSK licensed Chai Discovery's design models, but only after testing them in its own labs. Chai designed molecules for GSK's targets with no prior training on them, and GSK found working binders for every target it tested (Business Wire). Chai now has six big pharma partners in nine months (Endpoints News).

The business case is speed. Faster design-and-test cycles surface promising candidates sooner, and that matters in a race to market. Roche already says AI contributed to 40% of its pipeline decisions over the past three quarters (Reuters). One caution: no AI-designed molecule from these programs has proven itself in patients yet. The gains so far are at the front end of discovery.

What to watch: Who owns the official experiment record when software, not a scientist, starts the run. That choice affects data integrity rules in regulated labs. Also check whether your instrument data is standardized enough to feed a model at all.

Biohub pools $1.8 billion to build the data AI biology needs

Funders get the data a full year before everyone else.

On October 7, Biohub launched a $1.8 billion Virtual Biology Initiative to generate the large-scale cell data needed to train “virtual cell” models. Google DeepMind, Isomorphic Labs and Meta are putting in a combined $300 million. The U.S. Department of Energy and the National Institutes of Health (NIH) are taking part, and NVIDIA is a technology partner (Yahoo Finance). The first dataset is due in about a year. Companies that help pay for the data get a one-year head start on it before wider release.

What to watch: Whether any big pharma buys a seat to get inside that one-year window, and what the data license says about models trained on it.

Investors put a big price on AI drug discovery

Isomorphic's price tag comes before any of its AI-designed drugs has reached a patient.

On October 8, Bloomberg reported that Alphabet's Isomorphic Labs is in early talks to raise money at a valuation of at least $40 billion, and possibly $50 billion. Isomorphic plans to dose its first patients by the end of 2026, a year later than it first targeted (Startup Fortune). Meanwhile, NVIDIA-backed Iambic Therapeutics set terms for an initial public offering (IPO) at $15 to $17 a share, aiming to raise about $135 million (BioSpace). Highly valued AI partners will likely ask for more of your data in exchange for access to their models.

What to watch: Data rights and model ownership terms in your next AI partnership.

Illumina builds a stronger variant model into its clinical software

Better reading of genetic variants can change which patients qualify for a trial.

On October 8, Illumina released SpliceAI2. Illumina says it identified 17% more disease-relevant genetic variants than other splicing models in a rare disease dataset. It ships inside software that labs already use, including DRAGEN and Emedgene (Unite.AI). Separately, the European Medicines Agency (EMA) plans to update its 2018 guideline on good pharmacogenomic practice (RAPS).

What to watch: Whether your genomics pipelines record which model version produced each result, so past trial analyses can be reproduced.

⚖️ Regulatory & Policy

UK accepts all 44 AI recommendations and opens round three of its AI sandbox ↗

UK oversight of AI medical devices is shifting from one-time approval to monitoring for life.

On October 6, the UK government accepted every recommendation from its National Commission into the Regulation of AI in Healthcare (GOV.UK). The Medicines and Healthcare products Regulatory Agency (MHRA) will publish draft guidance by December on managing changes to AI devices after approval. It also opened Phase 3 of AI Airlock, its test program for AI medical devices, focused on monitoring after launch (GOV.UK). Applications for the current round close October 31.

What to watch: The December draft, and how it lines up with the U.S. Food and Drug Administration's (FDA) approach to pre-approved change plans.

FDA's new office for AI and real-world evidence takes shape

Drug sponsors may soon have one front door for AI questions in their submissions.

FDA's Center for Drug Evaluation and Research (CDER) says its new Office of Innovation and Clinical Trial Modernization has been operating since October 1 (FDAweb). On October 6, Clinical Trial Vanguard reported that it combines a Division of Artificial Intelligence and a Division of Real-World Evidence Analytics (Clinical Trial Vanguard). The notice that created the office did not spell out its functions, staffing or regulatory responsibilities, so what it will decide, versus advise, is still unclear.

What to watch: Whether the office reviews AI model credibility itself or only advises the review teams.

FDA plans to publish more of what companies used to keep private ↗

Clinical study reports and inspection findings may soon be public.

FDA posted its Transparency Roadmap on October 2, and trade press analyzed it this week. FDA plans to publish when drug applications are filed, release clinical study reports for approved drugs, and post redacted inspection findings, known as Form 483s (Applied Clinical Trials). It also notes a new agreement that lets FDA share nonpublic information with the Securities and Exchange Commission (SEC). On October 8, Eva Temkin of law firm Arnold & Porter warned that parts of the plan could draw legal challenges (RAPS).

What to watch: Whether your document systems can produce clean, redaction-ready versions of these records on short notice.

🔒 Cybersecurity & Risk

U.S. agencies expose a Chinese contractor behind state hacking

Manufacturing and university research partners are in the target zone.

On October 8, the Cybersecurity and Infrastructure Security Agency (CISA), the FBI and the National Security Agency (NSA) issued an advisory on Integrity Technology Group, a China-based contractor that supplies tools to state-backed hackers. The FBI seized web domains behind two of its hacking services (CyberScoop). Healthcare and manufacturing are listed as targets, and about 20 Taiwanese universities were confirmed victims (Nextgov).

What to watch: Phishing-resistant sign-in on Microsoft 365, and separating the research networks you share with universities.

FBI: attacks on Fortinet firewalls are still locking out administrators

Stolen firewall logins are a common first step for ransomware.

On October 7, the FBI warned that “FortiBleed” attacks on Fortinet firewalls and virtual private network (VPN) gateways are still underway (BleepingComputer). Attackers log in with leaked passwords, crack more offline, then lock the real administrators out. The FBI tied the campaign to ransomware groups. Researchers count about 86,600 compromised devices worldwide.

What to watch: Confirmation that every Fortinet device, including those at plants and contract manufacturers, is patched, has new passwords and uses multifactor authentication.

Unpatched flaw exposes in-house AI servers to takeover

Companies running their own AI models may have an open door.

On October 7, security firm JFrog disclosed a critical flaw in LMCache, open-source software that speeds up servers running large language models (The Hacker News). A single network message can take over the server, and no fix existed at disclosure. Many life sciences companies run AI in-house to keep trial and research data private, often on tools set up by data science teams outside IT's usual controls.

What to watch: Whether any internal AI platform uses LMCache, and blocking its network port until a patch ships.

Japanese drugmaker's doctor portal breach may affect 558,700 people

An outside contractor ran the site.

On October 6, Asahi Kasei Therapeutics disclosed a cyberattack on Pharma DIGITAL, its website for healthcare professionals. Data on up to about 558,700 people may have been exposed (BigGo Finance). The contractor that runs the site reported the breach on October 2, and the site was shut down. Portals like this are often built by agencies for marketing teams and sit outside central security monitoring.

What to watch: A list of every agency-run portal for doctors, patients and partners, and the breach-notice terms in each contract.

🎯 Leadership & Operating Model

ZS survey: 38% of pharma companies are getting real value from AI

Leaders invest in data first. Governance lags almost everywhere.

On October 6, ZS released its 2027 survey of pharma chief digital and information officers (CDIOs) (ZS). It classifies 38% of companies as “AI value leaders.” Measurable results in drug discovery doubled, from 17% to 35%. Leaders were far more likely to call data for AI mission-critical, 72% versus 46%. But only 17% say AI governance is fully working across the company.

What to watch: These benchmarks in your 2027 budget talks: fund the data layer before adding more AI projects.

Google gives AI agents their own identity and audit trail

AI agents are starting to look like staff that IT has to manage.

On October 8, Google launched a business AI agent that gets its own Workspace account and email address, with every action logged to the agent rather than a person (TechCrunch). Google says Merck is exploring Gemini Enterprise for molecular research (Google Cloud). Meanwhile, a Senate bill introduced October 1 would make developers and operators of AI agents liable for harm they cause (Yahoo Finance).

What to watch: Who approves an agent's access, who removes it, and whether it can be the named actor in a regulated audit trail.

Gilead renews Cognizant for five years and adds AI agents to software delivery

Outsourcing renewals are becoming the way AI enters IT itself.

On October 6, Cognizant announced a five-year extension with Gilead Sciences (BioSpace). It covers Gilead's software development and operations (DevOps) work and the use of AI agents in selected programs. Gilead's Global CIO, Anna Berg Asberg, framed it as practical, responsible AI (Pharmaceutical Technology).

What to watch: Who keeps the savings that AI produces, spelled out in your next outsourcing renewal.

Biopharma layoffs hit a low, but cost cutting now runs through technology

Pfizer is counting on technology for much of its next $2.5 billion in savings.

BioSpace's third-quarter count, published October 8, found 2,568 people affected by layoffs, the fewest since it began tracking in 2024 (BioSpace). Pfizer's added $2.5 billion in cuts relies heavily on “technology and simplification.”

What to watch: CFOs asking IT to show the savings from AI and automation.

💬 Joe's Take

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

Join the LS CIO Community →

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 before Joe adds his take.

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