GxPSoft AILife Sciences Software & AI — Week of October 5, 2026
Two weeks ago the story was that agents were shipping faster than the validation frameworks built to contain them. This week the government showed up and wrote the agents into the program design.
On September 30, 2026, HHS and ARPA-H launched SURPASS — Simulation-augmented, Real-time Platform Adaptive Seamless Trials — a five-year program whose third technical area is explicitly “an agentic operations layer” that automates trial startup, treatment-arm onboarding, and data collection, cleaning, and dataset construction. This is the first time a federally funded clinical research program has put autonomous operational software inside its architecture rather than on its wish list. The program also asks for publicly available validated standards — an invitation for this industry to contribute rather than inherit whatever gets built first.
Against that, CDRH published its FY2027 guidance agenda on October 1, and the prioritization is a story in itself: finalizing lifecycle management guidance for AI-enabled device software is an A-list priority, while “conducting risk assessments for generative AI-enabled devices” sits in the lowest tier. ValGenesis used a conference preview on October 2 to claim something we have been asking vendors for all year — measurable outcomes from AI-assisted validation in production pilots — without yet publishing the measurements. And Veeva renamed a field recording which AI produced a record: a small change with a real change-control footprint.
Here is what happened between September 29 and October 5, 2026.
ARPA-H Puts an Agentic Operations Layer Inside Federally Funded Clinical Trials
What happened: On September 30, 2026, the U.S. Department of Health and Human Services, through the Advanced Research Projects Agency for Health, launched SURPASS (Simulation-augmented, Real-time Platform Adaptive Seamless Trials) together with three complementary projects — STACK, COMMONS, and CINCH. The program targets a clinical development system that HHS describes as taking more than a decade, costing up to $2 billion on average, and failing more than 90 percent of the time.
SURPASS pursues three coordinated technical areas, quoted from the agency’s own release:
- A phaseless design engine — “integrating digital twins and other predictive models into trial design, simulating clinical and operational outcomes before launch, and developing the evidence needed to support regulatory confidence in these approaches.”
- A continuous inference engine — “always-valid, real-time or on-demand analysis, supporting rapid trial adaptations, and reducing the need for large conventional control groups.”
- An agentic operations layer — “automating key trial startup and operational activities, supporting onboarding of new treatment arms, and speeding data collection, cleaning, and dataset construction.”
The three satellite projects are infrastructure plays rather than trial-design plays: STACK uses AI to accelerate clinical site activation and “convert research-naïve sites into clinical trial sites”; COMMONS is a “privacy-by-design architecture for consent at national scale” for regulatory-grade data access; and CINCH helps patients contribute their own real-world data and find appropriate trials. Daria Fedyukina, Ph.D., is SURPASS program manager and oversees STACK and COMMONS; Erika Kim, Ph.D., oversees CINCH.
The program is an open funding opportunity. Solicitation ARPA-H-SOL-26-164 is posted to SAM.gov, with an informational webinar on October 15, 2026 (registration closes October 12), a Proposers’ Day on November 6 (registration closes October 28), a required solution summary due November 30, 2026, and proposals due January 22, 2027. No funding amount was disclosed.
“Today’s clinical trial system often requires too many stops, too much duplicated infrastructure, and too much time before researchers can understand whether a trial is on the right track,” Fedyukina said.
Why it matters: Read the third technical area as a data integrity mandate, because that is what it will become. If an agentic layer performs trial startup activities and data cleaning, it is producing and transforming records that feed a submission. Under 21 CFR Part 11 §11.10(e) those operations need secure, computer-generated, time-stamped audit trails; under ALCOA+, “Attributable” and “Contemporaneous” do not become optional because the actor was software. The genuinely new question is accountability: an agent that cleans a dataset is making a judgment, and someone named has to own that judgment in the trial master file.
The continuous inference engine is the second-order problem. “Always-valid” real-time analysis is a biostatistical property, but it is delivered by software, and the analysis software boundary — the thing you validate under GAMP 5 — now has to accommodate a model that updates estimates as data accumulate. If its outputs are interim decision inputs, its version, configuration, and change history are validation artifacts. The same holds for the digital twins in the phaseless design engine: FDA’s seven-step credibility framework expects a context-of-use statement and a model risk assessment for this model in this application, not a general platform claim.
Two things to look for before the proposal deadline: whether any team describes how the agentic operations layer is logged — what is written to the trial master file, in what format, with what human reviewer of record — and whether COMMONS’ consent architecture produces a provenance chain that survives an inspection.
Sources: HHS / ARPA-H — HHS launches SURPASS and new efforts to accelerate faster, smarter clinical trials, September 30, 2026; ARPA-H — SURPASS program page; STAT — HHS announces new efforts to speed up, expand clinical trials with AI, September 30, 2026; BioSpace — US government launches AI-driven programs to overhaul clinical trials, October 1, 2026
CDRH’s FY2027 Agenda: AI Lifecycle Guidance at the Top, genAI Risk Assessment at the Bottom
What happened: On October 1, 2026, FDA’s Center for Devices and Radiological Health published its FY2027 guidance agenda under MDUFA V — 11 planned final guidance documents and three draft guidances, split into an “A” priority list, a “B” list, and an “Under Construction” tier.
The top of the A-list is AI: CDRH’s stated priority is finalizing lifecycle management recommendations for AI-enabled device software functions, whose draft was issued in January 2025 and applies a total product lifecycle approach to AI-enabled devices. Finalizing Predetermined Change Control Plan guidance (draft issued August 2024) is also A-list, as is guidance on robotically assisted surgical devices. CDRH also plans a new draft on generative AI-enabled conversational devices for mental health disorders.
“Conducting risk assessments for generative AI-enabled devices” sits in the lowest-priority third tier, alongside predicate-device selection and thermal-effects evaluation, while drafts on software-function policy and post-market cybersecurity management sit on the B list. Comments on the prioritization are due November 30, 2026 (docket FDA-2012-N-1021).
Separately, the substance behind this agenda is the CDRH generative AI discussion paper released August 18, 2026, now in a comment period closing October 19, 2026. Per Cooley’s September 28 analysis, the paper proposes a two-axis risk framework — the degree and independence of the device’s activity against the severity of consequences if a user relies on an incorrect output — and treats measurement and signal-processing functions as higher risk because a user cannot independently assess the output. It floats a competency-based premarket model in which devices are assessed through benchmarking (safety, clinical proficiency, generalizability, plus “agentic-specific competencies” for agentic functions) followed by clinical confirmation scaled to risk, explicitly modeled on how clinicians are credentialed rather than on input-output testing. For postmarket, it contemplates three mechanisms: periodic device benchmarking on a defined cadence and after triggering events such as changes to the model or deployment architecture; periodic sample-based clinician review by qualified independent adjudicators; and performance degradation monitoring for drift.
Why it matters: CDRH has now described, in public and in detail, the most operationally specific postmarket regime for probabilistic software that any FDA center has put on paper. Even if you never touch a device, those three mechanisms are the template regulators are converging on: re-benchmark on model change, independent human adjudication on a sample, and continuous drift monitoring. Any GxP AI lifecycle plan lacking an analog to all three is going to look thin in 2027.
The prioritization is the uncomfortable part. Drift monitoring and periodic re-benchmarking are proposed for devices, while the drug-side pipeline has not caught up: CDER’s published guidance agenda lists a planned supplemental guidance on Computer Software Assurance for AI-Based Systems in Drug Manufacturing and Clinical Investigations, still a plan rather than a document as of this writing. So FDA is close to telling device makers how to keep an AI model honest over time, and has not yet told manufacturers how to keep a GMP model honest over time. That gap is where a great deal of unvalidated production AI currently lives.
One note on the tiering. Placing “risk assessments for generative AI-enabled devices” in the lowest-priority tier is defensible as bandwidth management and indefensible as a risk statement — tiers are publication schedules, but schedules get read as signals. The signal is that the agency does not intend to tell you how to risk-assess a genAI device before you have to file one.
Sources: RAPS — FDA’s device center releases guidance agenda for FY 2027, October 2, 2026 (announcement October 1); Cooley — Regulating AI Like a Doctor: FDA Floats Competency-Based Path for Generative AI-Enabled Devices, September 28, 2026; FDA CDRH — Proposed guidances for FY 2027; FDA — CDER guidance agenda (listing for computer software assurance for AI-based systems in drug manufacturing)
ValGenesis Claims Production Results for AI-Assisted Validation — Without the Numbers Yet
What happened: On October 2, 2026, ValGenesis announced it will present “real-world AI-assisted validation insights” at KENX Europe 2026: Harnessing the Power of AI in GxP, October 14–15, 2026, at the Four Elements Hotel Amsterdam. The company is a Platinum Sponsor.
The claims are architectural and methodological rather than a product launch. ValGenesis points to VAL™ (ValGenesis AI), “the governed AI natively built into the ValGenesis Smart GxP™ platform,” which “applies AI throughout the validation lifecycle while keeping humans in control of oversight, approval, and critical decisions.” The enumerated capabilities: generating validation documents, identifying potential compliance and test-coverage gaps, interpreting execution evidence, detecting anomalies, and analyzing document changes.
Kenneth Pierce, Ph.D., director of product growth, will lead two sessions. The first — “How Would You Know It Worked? Measuring AI-Assisted Validation in Practice” — draws on “customer pilot studies, implementation lessons, and measurable performance outcomes.” The second — “Pacing the Frontier: AI in GxP Life Sciences Operations” — “introduce[s] an AI maturity model for building trust in AI-supported processes and expanding adoption through clearly defined intended-use frameworks.”
“AI is moving beyond theory and into practical use in GxP environments,” said David Medina, chief marketing officer. “The challenge now is understanding where it adds meaningful value, how to evaluate the results, and how to maintain appropriate control.”
Why it matters: This is the closest a validation ISV has come in 2026 to promising the evidence class we have been asking for: not a capability list, but a measurement of whether the capability worked, drawn from customer pilots and production deployments. Two things about the framing are right. “Governed AI” placed inside a platform is at least an explicit architecture claim rather than a silent one. And “intended-use frameworks” is the correct vocabulary — the same term of art as in the joint FDA/EMA Guiding Principles for AI in drug development, and intended use is what determines whether an AI function is validation-relevant at all.
The gap is that “measurable performance outcomes” without published measurements is still a claim. The questions are operational. Measured against what denominator — per generated document, per test script, per executed protocol? Measured by whom — the vendor, the customer’s quality unit, or a third party? And is the pilot dataset available to a customer’s validation package, or only to a conference slide? An ISV that publishes a methodology its customers can re-run would be doing something genuinely new in this market; a conference session is a start, not a deliverable.
Sources: ValGenesis — ValGenesis to Bring Real-World AI-Assisted Validation Insights to KENX Europe 2026, October 2, 2026 (Business Wire); KENX — AI in GxP, Europe, October 14–15, 2026 — the Business Wire original returned an anti-bot block page this week; content read from the BioSpace syndication of the identical release.
Veeva Ships 26R2.3 With an AI-Provenance Field Migration
What happened: Veeva’s Q3 release cycle produced two customer-visible AI changes in the window, documented in release notes rather than in a press release.
Veeva Vault CRM 26R2.3 — release notes available October 1, 2026, sandbox October 8, production October 15 — carries a structural change to the agentic call-reporting path. The Agentic Call Report™ now uses a new Call Capture Agent in place of the Voice Agent. The data-model consequence is explicit: the ai_source__v field on the call2__v object is replaced by call_capture_agent_used__v, and ai_source__v “will be removed in a future release.” The switch and the agent’s permissions are automatically enabled with 26R2.3; “Admins do not need to make changes to configuration.” A second AI field appears in the same release: insight_quality_agent_ran__v on commercial_insight__v, an “Insight Quality Agent Ran” picklist that “capture[s] if the Insight Quality Agent ran on this record.”
Veeva Vault platform 26R2.3 — limited release October 2, 2026, general release December 4 — updates the Vault AI surface: the Query Agent returns “additional fields based on the object list view configuration” so follow-up questions need no second query, and has been tuned for faster responses and lower token use; the AI Tab gets progress UI for long-running agent operations and a record-detail retrieval tool that pulls multiple records “with a single call.” Veeva also shipped Document Translation via DeepL or Google Cloud Translate, with a Public Translation SDK usable “in workflow job steps and agent tools.”
Why it matters: Field renames are usually not news. This one is, for a Part 11 reason. A field named ai_source__v is an AI-provenance record: it is how the system says which AI touched a call record. Veeva is now asserting a narrower, more specific statement — call_capture_agent_used__v names the agent. That is a better data model, and it is also a change to a validated configuration that lands by default. If any report, integration, or downstream export reads ai_source__v, that dependency breaks; if your validation documentation lists the field inventory of call2__v, it is now stale. Neither event announces itself.
The auto-enablement is the warning we raised in September, sharpened. “Admins do not need to make changes to configuration” is true at the configuration layer and misleading at the control layer: in a Part 11 environment, a standing procedure to evaluate every limited release is the configuration change. The cadence — 26R2.2 on August 21, 26R2.3 on October 2, 26R2.4 on October 16, general release December 4 — means an unexamined release cycle carries an unassessed change roughly every four to six weeks. And if an investigator asks “which records were touched by an agent, and when,” that question is answerable only if you actually report on those fields — a report most Veeva customers have not built yet.
Sources: Veeva — What’s New in Vault CRM 26R2.3; Veeva — New in 26R3 Limited Release (Vault platform release notes, 26R2.3 / 26R2.4 schedule)
AbbVie Buys a Platform, Not an Asset: the Valkai Partnership
What happened: On September 29, 2026, AbbVie announced a partnership with Valkai, described as “a life sciences AI platform designed to help accelerate the pace at which scientific innovation reaches patients.” The scope, in AbbVie’s words: deploying “purpose-built AI capabilities designed to improve efficiency across targeted areas of clinical research, help teams generate insights more quickly and support opportunities to reduce development timelines.”
No financial terms, product names, deployment scope, timelines, or validation posture were disclosed. What the release does provide is framing. Valkai describes itself as “the AI operating system for life sciences,” with “Global Fortune 500 pharma and medtech leaders” as partners and a team drawn “from pioneering research and applied AI companies such as OpenAI, Google DeepMind, Sierra and Glean.” AbbVie situates the deal inside AI@AbbVie, its enterprise AI program, which it says is “guided by human expertise, responsible governance and a commitment to meaningful impact.”
“At AbbVie, we are focused on applying AI where it can create meaningful impact for our business and, ultimately, for patients,” said Nicholas Donoghoe, executive vice president and chief business and strategy officer. Andrew Campbell, M.D., vice president of clinical sciences, added that AI capabilities “have the potential to help us reduce time spent on manual processes and focus more attention on scientific interpretation and clinical strategy.”
Why it matters: The interesting fact is the shape of the deal, not its size. Set it against the 2026 pattern we documented in September: Merck × Google Cloud at as much as $1 billion, Roche × PathAI at $750 million upfront plus up to $300 million in milestones, Novartis × Orionis at $40 million upfront against up to $1.4 billion in milestones. Those were priced, milestone-bearing, asset-linked deals — pharma buying a model for a program. This one is unpriced and capability-linked: a pharma buying a general AI layer for clinical research. That is a different procurement motion, platform subscription rather than asset option, with a diligence path centered on the vendor rather than the model.
That is also where the GxP questions concentrate. A vendor-provided “AI operating system” spanning clinical development workflows is a supplier under your quality system and should be qualified against a defined intended use. AbbVie’s phrase “targeted areas” implies use-case-level scoping, the correct GAMP 5-compatible instinct — you qualify the system for the purpose, not the platform for its existence. The uncomfortable version of this deal, for which we have no evidence, is qualifying once for the platform and then onboarding workflows onto a moving target.
Three specifics we would want before the release becomes a template: what the audit trail looks like when Valkai’s agents touch clinical data; whether model or prompt changes flow through the customer’s change control or the vendor’s release notes; and whether the customer can export the evidence needed to defend an output two years later. “Responsible governance” is a commitment, not a control. Controls have version numbers.
MasterControl’s Two Gartner Categories — and the One Whose Definition Matters
What happened: On September 29, 2026, MasterControl announced it was named a Sample Vendor in two 2026 Gartner Hype Cycle reports: Hybrid Execution System, in the Gartner Hype Cycle for Life Science Manufacturing (published June 9, 2026), and Automated Compliance by Design, in the Gartner Hype Cycle for R&D New Product Development (published July 27, 2026).
The definitions are the substance, and MasterControl quotes them directly. A Hybrid Execution System is “an integrated platform that combines the functionalities of a manufacturing execution system (MES), quality management system (QMS) and a laboratory execution system (LES) to streamline and optimize both production and laboratory operations within an organization.” Automated compliance by design “embeds regulatory, safety and quality requirements as enforceable constraints within new product development workflows, using automation and traceability to continuously evaluate compliance as designs change.”
“We believe being named a Sample Vendor in two Gartner Hype Cycle reports in the same year reflects the range of problems our platform is built to solve,” said Dave Edwards, chief executive officer. MasterControl describes itself as “the premier AI-fueled quality and manufacturing platform for life sciences.”
Why it matters: Two category names, two very different levels of maturity — and one of them is going to be abused.
Watch “automated compliance by design.” Read the definition carefully and it describes a rule engine, not a language model: requirements become “enforceable constraints,” and compliance is “continuously evaluate[d]… as designs change.” That is deterministic logic with version control, and deterministic logic is auditable. The term is now in a Gartner Hype Cycle, which means within two quarters several vendors will claim it for products where the “enforcement” is a model-mediated suggestion the user can dismiss. When you evaluate such a claim, ask one question: is the compliance rule set versioned, diffable, and independently testable? If yes, it is a control. If no, it is advice, and advice does not survive an inspection.
“Hybrid execution system” is the more consequential category for validation architecture. Merging MES, QMS, and LES collapses three historically separate validated systems and three change-control surfaces into one. That can be a genuine improvement — a single traceable chain from batch record to deviation to lab result is what data integrity wants — but it also means one supplier’s release train now moves your validated boundary in manufacturing, quality, and the lab simultaneously. Before the category becomes a purchasing criterion, ask for the validated-instance story: one validated platform, or three logical systems with independent release and change control? GAMP 5’s system boundaries exist for a reason; vendors who merge them owe customers an explicit answer about the seams.
Standard caveat: Hype Cycle placement reflects analyst opinion, and Gartner “does not endorse any company, vendor, product or service depicted in its publications.” A Sample Vendor listing is a market-map data point, not a validation.
The Rest of the Tape: Awards, Diagnostics, and the Manufacturing Edge
Not every item deserves its own section. Several share one theme: the deployment surface for real AI in regulated life sciences this week was the floor and the bench, not the enterprise workflow.
PubHive was named “Best Life Sciences Workflow SaaS Platform 2026” in The SaaS-ies, presented by Acquisition International, in a release dated September 30 and published October 2. PubHive Navigator connects literature monitoring, pharmacovigilance workflows (ICSR, signal, aggregate reporting), medical and clinical evidence management, and regulatory activities. The listed core area worth noting is “Human-in-the-Loop AI-Assisted Review: Applying intelligent assistance while keeping scientific professionals responsible for review and decision-making” — the correct posture for AI-assisted literature surveillance, where the record you produce is safety-relevant. It is the company’s second award announcement in three weeks, and awards are marketing; the architecture statement is the part with substance.
Caris Life Sciences unveiled an enhanced Caris Assure on September 29, 2026: a liquid biopsy integrating whole exome sequencing, whole transcriptome sequencing, and whole-blood analysis for clonal hematopoiesis identification in one blood draw, with vendor-stated coverage across more than 23,000 genes and a median turnaround of about seven days. CareDx drew September 29 coverage for AI-driven transplant surveillance and Massive Bio was named to TIME’s World’s Top HealthTech Companies 2026 — both headline-level items, and neither a product milestone we can verify independently.
And the manufacturing floor, at CPHI Milan 2026 (October 6–8), gets an AI and tech zone for the first time. Two pre-show announcements point at where vision AI is actually landing. Steriline is showing its VKSFCM11 robotic filling line with the OmniAI Vision system and the OptiLAIN predictive-maintenance platform; the engineering rationale in the company’s own words is that cameras sit away from the filling, capping, and crimping area to respect the “first air” concept while widening the field of view. Systech announced the ST15S semi-automated serialization and aggregation station and the ST16P multi-aggregation and automated label print-and-apply station on September 30, both running on its UniSeries software.
Steriline deserves a line of its own because it points at the most under-discussed validation topic in the industry. A vision system on a sterile filling line is an inspection function in a GMP-critical process, and vision models drift with lighting, product presentation, and camera maintenance in ways an annual revalidation does not catch. Annex 1 gave contamination control a rigorous framework; nobody has yet given machine vision on a filling line an equivalent — performance monitoring, periodic re-qualification after hardware intervention, and a documented drift policy. Put those three items in the qualification protocol now, because the vendor’s datasheet will not.
Upcoming Events
- KENX AI in GxP Europe — Amsterdam, October 14–15, 2026. Two days on leveraging and validating AI in GxP environments. ValGenesis is a Platinum Sponsor; Kenneth Pierce’s session on measuring AI-assisted validation performance is the one on the agenda we would attend.
- ARPA-H SURPASS informational webinar — October 15 (register by October 12). Proposers’ Day follows November 6 (register by October 28); solution summaries are due November 30 and required to advance; proposals are due January 22, 2027.
- CDRH genAI discussion paper — comments due October 19, 2026. The two-axis risk framework and the competency-based premarket model are both open for input. This is the highest-leverage comment window in this digest.
- Veeva Vault CRM 26R2.3 production release — October 15, 2026. Check whether
ai_source__vdeprecation breaks reports or integrations. Vault platform 26R2.4 lands October 16; general release December 4. - CDRH FY2027 guidance prioritization — comments due November 30, 2026 (docket FDA-2012-N-1021). If genAI risk assessment sitting in the lowest tier bothers you, this is the mechanism to say so.
- CPHI Milan 2026 — October 6–8. First AI and tech zone, alongside cold chain, contamination control, and labeling.
- Veeva R&D and Quality Summit — Boston, October 20–21, 2026. The first venue likely to carry detail on the Falcon Agent Suite beyond the press releases.
- EU GMP Annex 22 status. Draft published July 7, 2025; consultation closed October 7, 2025; EMA’s work plan targets Q4 2026 for final text to the European Commission, and as of early September it remained in draft. Watch for adoption — it would be the first binding AI-specific GMP text anywhere.
What to Watch Next Week
- Does anyone describe how the SURPASS agentic operations layer is logged? The tell will be in the proposal language, not the launch release. A program that asks for auditable agent activity will produce reusable architecture. One that does not will produce a demo.
- Whether ValGenesis publishes the measurement methodology behind “How Would You Know It Worked?” A conference session in Amsterdam on October 14 is the moment. If the numbers come with a denominator and a re-runnable method, that is a genuinely new contribution to validation practice. If they come as percentages on a slide, it is a capability claim with better typography.
- Veeva’s
ai_source__vdeprecation in the field. Expect at least one customer to discover a broken report. Expect nobody to file it as a change-control event. The second one is the problem. - “Automated compliance by design” as a term. It is now a Gartner category. Count how many vendors claim it within a quarter, and ask each for the versioned rule set.
- The hybrid execution system boundary problem. Watch whether MasterControl or any competitor publishes an explicit statement about validated instances and change control when MES, QMS, and LES share a platform. That answer is worth more than any Hype Cycle listing.
- Whether the FDA/EMA Guiding Principles get teeth. Annex 22 is the vehicle: a principles document tells you what good looks like, an annex tells you what an inspector will cite. The gap between them is currently the entire regulatory exposure of AI in regulated operations.
Source Notes and Caveats
We mark every claim with its source. Where we could not reach a primary document this week, we say so rather than reconstruct it:
- Business Wire was anti-bot gated again. The ValGenesis release returned a block page; we read the identical text from the BioSpace syndication, which carries the Business Wire attribution.
- Two items that looked like news were re-syndications. Dotmatics’ Luma Agent was announced May 13, 2026 and Sapio Sciences’ Claude Cowork integration on April 29, 2026, both appearing in searches with fresher third-party dates. Also date-checked and excluded: Roche × Dualitas (September 17, 2026) and Ginkgo Bioworks’ autonomous lab at Novo Nordisk’s Waltham site (September 15, 2026).
- All performance and efficiency figures are vendor-supplied and were not independently verified in this session — including ValGenesis’ pilot-outcome claims, Caris’ sequencing figures, CareDx’s risk-prediction framing, PubHive’s workflow claims, and Steriline’s and Systech’s automation claims.
- Regulatory detail is second-hand where noted. The CDRH FY2027 agenda items are read from coverage of the October 1 announcement and the agency’s guidance listings, not from the guidances themselves, none of which are final. The genAI discussion paper details come from Cooley’s September 28 alert; we did not separately re-read the underlying FDA paper. The CDER guidance-agenda item on computer software assurance for AI-based systems in drug manufacturing was read from the agenda listing, not an issued guidance — it is a planned document.
- Gartner Hype Cycle placement is analyst opinion, and Gartner disclaims endorsement of the vendors depicted. Veeva release dates are “subject to change” in Veeva’s own notes. The September 2026 deal-table figures in the AbbVie section are carried forward from our September 28 edition and were not re-verified.
- No funding amounts were disclosed for ARPA-H SURPASS, and we are not estimating one.
- Quiet window: no fresh AI announcements were located this week from Benchling, IQVIA, Qualio, Greenlight Guru, ComplianceQuest, ETQ/Hexagon, Dot Compliance, Sapio Sciences, LabVantage, Thermo Fisher, L7 Informatics, Kneat, Dassault Systèmes/ArisGlobal, or Dotmatics. Absence here means absence of announcements, not of activity.
Published by GxPSoft AI — Architecture and compliance intelligence for regulated life sciences software. We track the products, not the press releases.
Related reading from our archive: Life Sciences Software & AI — Week of September 28, 2026 · Current State of AI in Quality Management Vendors — A 2026 Field Guide · The Validation Data Fabric: A Three-Layer Architecture · How QMS Vendors Ship AI in Part 11