|
Tight 6–8 minute read, with source links for deeper exploration.
Research conducted and compiled with Perplexity and Claude AI.
Welcome back. This week, three forces collided in ways that demand your attention: AI clinical development partnerships reached a tipping point — Evinova locked in simultaneous deals with AstraZeneca, Astellas, and BMS in a single week — while the institutions meant to protect your digital infrastructure hit new lows, with CISA furloughing 60% of its remaining staff and replacing its acting director mid-crisis. A ransomware attack that last week shuttered 36 UMMC clinics moved into recovery, delivering the first hard benchmark on what a 10-day healthcare system outage actually costs operationally. And new survey data from Benchling, McKinsey, and Kiteworks converged on a single uncomfortable truth: organizations are deploying AI and autonomous security tools faster than they are governing them — and the regulatory window to fix that before it becomes a compliance problem is closing.
💡 Leadership & Operating Model content is at the bottom this week — if that's where you want to start, jump there now.
🤖 AI & Data
Evinova Locks In Three Top-Pharma Deals in One Week — and the Data-Sharing Model Is the Real Story
Between February 18–23, Evinova — the AI-native clinical development company spun out of AstraZeneca in 2023 — formalized simultaneous collaborations with AstraZeneca, Astellas Pharma, and Bristol Myers Squibb. The structural differentiator isn't any single deal — it's the collective data-sharing model underneath all three.
What happened:
- All three sponsors feed anonymized operational trial data back into Evinova's platform — a network effect that improves AI recommendations as more partners join
- Platform spans study design, regulatory authoring, and CDISC USDM 4.0 digitization across multi-agent AI capabilities
- First published efficiency benchmarks: 5–7% savings per study, translating to hundreds of millions of dollars across a top-10 pharma portfolio
- AZ CEO Pascal Soriot cited faster trials and better patient experience; Astellas CRDO Tadaaki Taniguchi cited smarter study design and reduced rework
Why it matters to you:
- Early adopters are shaping the platform's data model and benchmarks — late joiners will inherit terms they didn't set
- This is the reference architecture for AI-native clinical platform partnerships; your clinical operations team needs to evaluate integration compatibility now
- The multi-sponsor model creates compounding AI improvement — organizations outside the network fall further behind with each new partner that joins
📋 What to Watch: Define — before it becomes a contract negotiation — what anonymized operational trial data your organization will contribute to shared AI platforms, under what governance terms, and with what data residency requirements.
Benchling's 2026 Biotech AI Report: Data Infrastructure Is the Bottleneck, Not the Models
Drug Discovery News synthesized the Benchling 2026 Biotech AI Report on February 23, surveying ~100 biotech/pharma organizations actively using AI in R&D.
Key findings:
- 80% plan to increase AI budgets in the next 12 months; 23% plan to double spend or more
- 55% cite poor data quality — not model performance, not vendor capability — as the primary reason pilots fail
- Adoption highest where data is clean: protein structure prediction (73%), literature extraction (76%)
- Adoption drops in complex domains: generative design (42%), biomarker analysis (40%), ADME prediction (29%)
- 67% prioritize internal upskilling over external tech hiring — the "scientific translator" (biology + ML + regulatory) is the critical emerging hire
- High-AI-adoption organizations are nearly 2x as likely to report strong wet-dry lab integration (30% vs. 18%)
Why it matters to you:
- Read alongside last week's ZS finding (only 40% of pharma AI pilots scale), two independent data sets now point to the same root cause: infrastructure and integration gaps, not model limitations
- If informatics and lab operations teams aren't jointly accountable for data quality, the 55% failure rate is your forecast, not the exception
📋 What to Watch: Benchmark your wet-dry lab integration maturity against this report. The integration gap is where AI ROI is being lost — not in the model selection decision.
Sanofi's "Plai" Is the First Public Proof That Agentic AI Is Inside Pharma Governance — Not Just Operations
On February 23, Sanofi disclosed operational details of "Plai," its agentic AI portfolio decision tool built with Aily Labs — and the detail that matters is where in the governance process it sits.
What happened:
- Plai aggregates 1 billion+ data points to predict R&D costs, trial enrollment timelines, and probability of success
- Drug development committee meetings now begin with Plai's AI-agent recommendation on program advancement — human override is possible, but AI sets the starting point
- CEO Paul Hudson's February 10 Fortune interview added: 10 new drug targets identified in one year via ML; AI supply chain tools have avoided $300M in revenue risk
Why it matters to you:
- This is the most concrete public example of a major pharma company embedding agentic AI directly into regulated governance decisions — not back-office productivity
- The moment a governance body opens its meeting with an AI agent's recommendation, you have an audit trail obligation, an explainability requirement, and potentially a regulatory disclosure question
- Boards and regulators will begin asking about this architecture — your answer should exist before they ask
📋 What to Watch: When your governance bodies begin using AI-agent input as a starting point for drug program decisions, what documentation, explainability, and regulatory disclosure obligations attach? Design that answer before you deploy the tool, not after.
⚖️ Regulatory & Policy
Arnold & Porter's EU Digital Omnibus Analysis: Extended Timelines, Integrated Pathways — and New Health Data Rights
On February 25, Arnold & Porter published the most comprehensive legal analysis to date of the EU Digital Omnibus proposals — two November 2025 legislative packages amending the EU AI Act, GDPR, ePrivacy Directive, NIS2, and EU Data Act simultaneously.
Four changes that matter most for life sciences CIOs:
- Extended compliance runway: AI Act high-risk obligations for Annex I systems (medical devices) now tied to harmonized standards availability — potentially pushing to August 2028
- Integrated conformity assessment: Single pathway proposed for EU AI Act + MDR/IVDR — reducing duplicative compliance work for AI-enabled devices and SaMD
- New health data legal basis: Explicit authorization to process special category health data for AI bias detection and correction
- Expanded real-world testing and sandboxes: Extended to Annex I products; EU-level AI sandboxes available through the AI Office
Status: The European Parliament has begun publishing its own amendments — final scope is still in motion.
📋 What to Watch: The integrated conformity assessment pathway is the most operationally significant change for medical device CIOs. Inventory all AI-enabled devices and SaMD in your EU portfolio now, and model what the integrated pathway means for existing MDR compliance timelines before the Omnibus reaches final vote.
CISA's Acting Director Is Out, 60% of Staff Are Furloughed, and the CIRCIA Clock Is Still Running
Two developments materially changed the CISA picture from last week's coverage.
What moved this week (new vs. last week):
- Director replaced: On February 27, DHS reassigned acting director Madhu Gottumukkala and installed Nick Andersen — following reports of sensitive documents uploaded to a public ChatGPT instance and a failed counterintelligence polygraph
- Operational degradation confirmed: A partial DHS shutdown beginning February 14–15 has furloughed ~60% of CISA's ~2,400 remaining staff; the NYT reported February 22 that long-term planning, training, prevention work, and new contracts have all paused; CISA CIO Bob Costello also departed
- CIRCIA scope signal — NEW this week: Covington/Inside Privacy (Feb 23) and Crowell & Moring (Feb 18) both confirm CISA is explicitly signaling willingness to narrow the rule's scope and harmonize with existing HHS/OCR requirements — a shift not present in last week's coverage
Key date: Healthcare & Public Health CIRCIA sector town hall — March 17
📋 What to Watch: CISA's degraded capacity makes the March 17 session more important, not less. Prepare specific, data-backed comments on harmonizing existing HHS/OCR reporting obligations with CIRCIA. The scope-narrowing signal means industry input can still move covered entity definitions for pharma manufacturers, CDMOs, and CROs — but that window will not stay open.
🔒 Cybersecurity & Risk
UMMC: Clinics Reopen After 10 Days — But the Full Reckoning Will Take Months
Last week we covered the February 19 ransomware attack that shut down all 36 UMMC clinics and took Epic offline. This week delivered the recovery timeline data.
Recovery milestones:
- February 28: Epic/MyChart restored
- March 2: All statewide clinics reopen — 10-day closure total
- Extended clinic hours planned to absorb appointment backlog
- Mississippi Today quoted experts on February 24 estimating "weeks to months" for complete system restoration
- Vice Chancellor LouAnn Woodward: "It will take some time to fully recover and to investigate this criminal attack"
What remains open:
- FBI investigation into attacker identity and data exfiltration is active
- UMMC has not confirmed whether a ransom was paid or patient data was compromised
Why it matters to you:
- The 10-day clinic closure is now a concrete industry benchmark for a mid-tier ransomware event
- Academic medical centers are frequent clinical trial sites, real-world data partners, and specialty pharmacy hubs — their resilience is your resilience
- "Weeks to months" for full recovery means residual risk persists well past the reopen date
📋 What to Watch: Map your academic medical center dependencies — trial sites, data pipelines, patient support operations — and test your data continuity plan specifically against a 10-day partner outage. That is now the documented baseline, not a worst-case estimate.
BeyondTrust CVE-2026-1731: CVSS 9.9, Healthcare Targeted, Active Exploitation Confirmed
A critical flaw in BeyondTrust's Remote Support (RS) and Privileged Remote Access (PRA) products — widely used in life sciences for vendor remote access and privileged session management — has been actively exploited since February 10.
The vulnerability:
- CVE-2026-1731, CVSS 9.9 — unauthenticated RCE via a single crafted WebSocket message; no credentials, no user interaction, low complexity
- CISA added to KEV catalog: February 13
- Orca Security identified 16,400+ internet-exposed instances globally
Confirmed post-exploitation activity (Palo Alto Unit 42, Feb 23):
- Domain account creation and lateral movement
- Backdoor deployment (VShell, SparkRAT)
- Credential vault exfiltration
Patch status:
- SaaS customers: patched by BeyondTrust February 2
- Self-hosted: upgrade to version 25.3.2 or later immediately
- AHA/Health-ISAC issued a TLP:WHITE healthcare-targeted bulletin February 19–20
📋 What to Watch: Patch all self-hosted instances to 25.3.2+ now. Audit access logs back to January 31. Hunt for VShell/SparkRAT IoCs and unauthorized domain accounts created in the past 30 days. BeyondTrust PRA/RS is a common CRO and equipment vendor access pathway — that's exactly what attackers are exploiting.
Marquis v. SonicWall: When Your Vendor's Breach Becomes Your Breach — and Your Lawsuit
On February 23, Marquis Software Solutions filed suit against SonicWall in the Eastern District of Texas — establishing what Dark Reading called on February 27 a potential watershed for vendor cybersecurity liability.
What happened:
- A breach of SonicWall's MySonicWall cloud backup service exposed firewall configurations, VPN setups, and MFA scratch codes for all customers
- Attackers bypassed Marquis's fully patched, MFA-enabled defenses entirely using the vendor's stolen configuration data
- Mandiant attributed the attack to state-sponsored actors; 400,000+ individuals affected across 700+ financial institution clients
- Marquis now faces 36+ class action lawsuits from its own downstream customers
The legal argument:
- No amount of customer-side controls can compensate for a vendor's failure to secure its own infrastructure
- The vendor's breach caused downstream customer harm — and the vendor should bear liability
📋 What to Watch: Audit whether firewall, VPN, or security vendors store configuration backups in their cloud — and whether those backups include MFA recovery codes. Review vendor contracts for liability language when a vendor's own breach enables a downstream attack on you. This case will set a precedent that reshapes third-party risk expectations across regulated industries.
💡 Leadership & Operating Model
Two frameworks published this week reframe the CIO's core mandate in 2026: the job is no longer managing technology — it is architecting the governance layer through which humans and AI agents jointly produce business outcomes. In life sciences, that framing carries weight that most other industries don't face: every AI-generated output touching a regulated process carries documentation, explainability, and oversight requirements that must be designed in, not bolted on.
McKinsey's Global Tech Agenda 2026: CIOs Are Strategy Architects — and the Data Gap Is Structural
Published February 8, McKinsey's Global Tech Agenda 2026 documents a measurable divide between top-performing CIOs and the rest.
The divide:
- ~65% of top performers actively involve their CIO in enterprise strategy vs. 52% at others
- 28% of top performers plan budget increases exceeding 10% vs. 3% at others
McKinsey's four imperatives for 2026:
- Deploy agentic AI for end-to-end process automation — not isolated pilots
- Build genuine in-house technology capability, not perpetual vendor dependency
- Replace annual planning cycles with continuous, nimble decision-making
- Build talent models combining engineering excellence with domain expertise
The honest caveats (even among top performers):
- 1 in 4 lack data foundations required for agentic AI
- Nearly 1 in 3 struggle with AI talent gaps
- System integration remains the most persistent delivery blocker
📋 What to Watch: Map your position against these four imperatives — specifically whether your planning cycle, talent model, and data architecture are calibrated for the speed at which agentic AI decisions are now being made at peer organizations. If your data architecture can't support multi-agent orchestration, the agentic future isn't just delayed — it is structurally blocked.
The CIO as Chief Intelligence Orchestrator: Agent Governance Is the New Core Mandate
CIO.com's January 25 feature — which accumulated significant C-suite traction through late February — crystallizes the emerging consensus: the CIO role is evolving from infrastructure steward to orchestrator of a unified workforce spanning human employees, AI copilots, and autonomous agents.
Key signals:
- Saket Srivastava (CIO, Asana): "CIOs will be judged on how effectively they improve cycle time, predictability, and throughput in the workflows that drive revenue"
- 82% of CIOs view themselves as "unifiers advocating for innovation across silos" (Forbes research)
- Other C-suite executives name the CIO as their most desired collaboration partner
- Gartner projects CIOs will soon manage agentic HR, own strategy and P&L outcomes, and govern portfolios extending beyond IT
The life sciences-specific governance question:
- Who in your organization has authority to deploy an AI agent into a regulated workflow?
- What change-control, audit trail, and documentation requirements apply when they do?
- Most life sciences organizations have not formally answered either question
📋 What to Watch: If your organization lacks a formal policy governing who can create, deploy, modify, and retire AI agents — and what documentation applies when those agents touch regulated processes — that gap will become a compliance finding before it becomes a strategic priority. Build the governance protocol now, while you set the terms.
The 77/37 Governance Gap: AI Is Running in Your Security Stack Without a Policy
The Kiteworks State of AI Cybersecurity 2026 report (~February 24) quantified a cross-industry governance failure that lands directly on the CIO's desk.
The numbers:
- 77% of organizations have deployed GenAI or LLMs in their cybersecurity stack
- Only 37% have a formal AI policy governing that use
- 73% report AI-powered threats are already impacting their organization
- 18% of executives believe AI operates with high autonomy in their SOC — vs. 14% of practitioners (leaders are overestimating their AI maturity)
Combined with last week's Censinet data (70% have AI governance committees, only 30% have AI inventories):
- Two independent data sets, same pattern: governance structures built before foundational inventory work is done
- Committees are governing systems they haven't catalogued
- Board-level AI reporting is structurally detached from operational reality
📋 What to Watch: Ask your CISO to enumerate every AI/ML tool deployed in your security stack with data access scope and governance classification for each. If the answer is "we don't have that list," you have autonomous systems operating inside your security perimeter without documented oversight — and that needs to exist before your next regulatory examination, not after.
✏️ Editor's Perspective
What strikes me most about this week's developments is how consistently the governance question surfaces regardless of which domain you're in. Evinova's triple-pharma data-sharing model is an AI story — but the strategic leverage sits entirely in the governance architecture: who sets the terms of data contribution, who owns the benchmarks produced, and what happens to the AI's recommendations when one sponsor's dataset dwarfs another's. Sanofi's Plai disclosure is an innovation story — but the moment a drug program committee opens its meeting with an AI agent's recommendation, you have an audit trail problem, an explainability obligation, and potentially a regulatory disclosure question that didn't exist before. The UMMC recovery timeline and the Marquis v. SonicWall lawsuit are cybersecurity stories — but they're really accountability stories about what happens when governance assumptions turn out to be untested rather than verified.
The throughline this week is the same as last week, reinforced from multiple directions: the CIOs pulling ahead are treating governance as operating infrastructure — not as a compliance requirement that follows capability deployment. They're defining data-sharing terms before joining platforms. They're auditing vendor backup environments, not just vendor SLAs. They're inventorying AI agents before standing up governance committees. The organizations that do that work now will have a structural advantage when regulatory frameworks harden. The ones that don't will be doing remediation under deadline.
🔗 Top 3 Must-Read Links This Week
- Evinova Triple-Pharma Partnership — FierceBiotech (Feb 23) — Read the primary source for the multi-sponsor data-sharing architecture details that will define AI-native clinical platform partnerships across the sector.
- Arnold & Porter EU Digital Omnibus Advisory (Feb 25) — The most operationally specific legal analysis of the EU AI Act/MDR/GDPR convergence currently available; essential reading for any CIO with EU medical device or clinical AI programs.
- McKinsey Global Tech Agenda 2026 — Primary source on CIOs as strategy architects, with the top-performer vs. laggard data that belongs in your next board briefing.
Ready to go beyond the headlines? Join the conversation in the Leadership Inklings community — where life sciences CIOs and IT leaders connect, share what's actually working, and build on intelligence like this together.