Bristol Myers Squibb

Ortent Life Sciences Radar, research beta. Evidence to 6 October 2026, rule set v0.4.3. Scope: AI and computation in drug discovery (T1), and research-lab digitalisation and scientific data systems (T2). Company scope: the whole company, as the cohort 2 scope notes set it. Each ID in brackets links to its online source, and the source index at the end of this brief lists them.

Disclosure: Andrew Wyatt, Ortent's principal, was Chief Growth Officer at Sapio Sciences until April 2026. Software vendors, including Sapio and its competitors, appear in this brief only where public sources name them.

Status: research beta. Ortent has reviewed this brief; it has not been independently validated. It summarises activity identified in public sources. Search coverage varies by company and is set out under Coverage limits, and zero identified items does not mean no activity. It does not predict buying intent or measure operational maturity. For this radar, process and CMC development are grouped with manufacturing and excluded from the research-lab theme (T2); such work appears as context where it helps.

Summary

The reviewed sources identify six T1 initiatives with an AI-discovery commitment change involving Bristol Myers Squibb (BMS) first disclosed in the year to 5 October 2026, and three in the year before [>BMS-E02, >E11, >E12, >E14, >E15, >E19, >E21, >E24, >E25]. They include an expansion of BMS's NVIDIA compute to a DGX SuperPOD with Vera Rubin systems, which BMS announced itself, and partner-announced or media-reported agreements with Chai Discovery and Schrödinger [>BMS-E02, >E25, >E24]. We identified no commitment change in company lab digitalisation (T2) in either year in the searched sources. A dedicated T2 search run on 8 October 2026, after the evidence cutoff, found no further eligible BMS disclosure. The 6 October recruitment snapshot contains 6 relevant R&D requisition postings across the two themes, none with ambiguous vacancy status [>BMS-R02, >R17, >R25, >R26, >R27, >R39]. These observations identify programmes and functions for further investigation; they do not establish procurement opportunity, deployment completion or net hiring.

Announced commitments

BMS published one of these nine rows itself [>BMS-E02] and issued two jointly, with Orbital and with insitro [>BMS-E19, >E21]. The OpenFold row rests on a consortium news page that links to a Business Wire release we could not read [>BMS-E11]. Two rows are Apheris releases that quote a BMS executive [>BMS-E12, >E14], and two are partner releases with no BMS quote, from Evotec and Chai Discovery [>BMS-E15, >E25]. The Schrödinger row rests on a trade press article that quotes a BMS vice president; the Schrödinger release was not reached [>BMS-E24]. We found no BMS confirmation of these six rows. The BMS newsroom was blocked and the IR feed omits some deal releases, so any confirmation published only on the newsroom was not reachable.

The Evotec row counts as an expansion because the release states new scope and a US$50 million payment. The payment is described as "programme-based", so it could instead mark a pre-agreed milestone; read that way, T1 W0 would be two [>BMS-E15]. The insitro row is an expansion because the joint release announces "the next phase of their collaboration" with new funding [>BMS-E21]. Orbital counts in T1 under C42-2 because the release states that BMS will acquire the RNA platform with AI-driven design, not only the lead programme [>BMS-E19]. The Schrödinger article does not date the agreement, so that row counts by first disclosure only [>BMS-E24].

By event date, we counted three T1 initiatives with a qualifying commitment change in W0 (5 October 2024 to 4 October 2025): Evotec, OpenFold and the Federated OpenFold3 initiative. We counted five in W1 (5 October 2025 to 5 October 2026): Orbital, insitro, the NVIDIA Vera Rubin expansion, Chai and AISB Bind. By first disclosure, the T1 count is three in W0 and six in W1, adding Schrödinger, whose event date is uncertain. T2 is zero in both views and both years. One T1 milestone was disclosed in W0 and six in W1 [>BMS-E16; >E13, >E17, >E18, >E20, >E22, >E23]. Existing capabilities first disclosed numbered three in T1 and one in T2 in W0 [>BMS-E05, >E08, >E09; >E26], and five in T1 and none in T2 in W1 [>BMS-E01, >E03, >E04, >E06, >E07]. These are not counted as changes. The Rescale modelling item, an existing T2 capability in W1 under rule set v0.4.2, is process development and is recorded as context under the v0.4.3 theme 2 boundary [>BMS-E27]. Under the opposite reading of borderline clarification C42-2, counting AI-discovered in-licensed assets as T1 would make the reported discontinuation of an Exscientia-discovered inhibitor a candidate T1 negative change and raise T1 W1 by first disclosure to seven, while treating Orbital as an asset deal would make T1 W1 four by event date and five by first disclosure [>BMS-E47, >E19]. Two other readings would change T1 W1: counting the Anthropic agreement as T1 would add one initiative in both views, and taking the Schrödinger article date as the event date would make it six by event date [>BMS-E28, >E24].

Coverage is partial for both themes in both years. The annual report and 10-K, consortium, academic and negative-event search families are complete, and the per-event-type web search is complete for T1 but partial for T2. The company press release archive, results transcripts, partner newswires and venture families are partial. The table marks the partner newswire and per-event-type search families as not comparable between the years, so no comparison between the years is made.

Context items, including the Anthropic enterprise agreement, are recorded separately (see Context).

Identified recruitment

On 6 October 2026 jobs.bms.com listed 676 unique requisitions by one reviewer's count and 680 by the other's. Both reviewers applied a common title filter, which returned the same 37 requisitions, and the final table classifies 45 postings from their duties. We observed 6 relevant requisition postings in the R&D theme class, 4 T1 and 2 T2 [>BMS-R02, >R25, >R26, >R27; >R17, >R39]. No requisition is tagged to both themes. No BMS posting shows an openings field, so vacancy status is not recorded for any of the 6 and none has ambiguous vacancy status. A further 19 are enterprise roles, 9 are plant, quality or manufacturing roles, 1 is an internship and 10 are out of scope. These are requisition postings listed on one day, not hires, headcount or a trend.

The T1 postings describe AI and computation roles in discovery: - in Research Computational Sciences, an Associate Director to lead the Machine Learning Sciences team, which works on ADMET and PK prediction and multiparameter design for small molecules [>BMS-R02] - a Principal Scientist in the Small Molecule Computational Design team in San Diego, using computational chemistry, cheminformatics and machine learning to design degraders and molecular glues [>BMS-R26] - in Informatics and Predictive Sciences (IPS), a Principal Scientist in computational biology in Cambridge, Massachusetts, to apply machine learning to perturbation screens and omics data and help "nominate and validate new targets" [>BMS-R27] - in IPS, a remote Principal Research AI Innovation Lead to build agentic LLM systems for research workflows, including "target evidence assembly" [>BMS-R25]

The T2 postings describe lab data and lab automation roles: - a Discovery Research Domain Analyst in the Molecular Invention Data Product Team, in Princeton and Cambridge, Massachusetts, to document scientific data workflows and maintain metadata that supports "FAIR-aligned practices" [>BMS-R17] - a Senior Scientist in Discovery Biotherapeutics, Brisbane, California, to lead high-throughput biologics production with laboratory automation and to work with automation, data and computational teams to "advance closed-loop discovery workflows" [>BMS-R39]

A PD Senior Engineer in Hyderabad, for model lifecycle and MLOps in the Pharmaceutical Product Development Model Hub, which holds manufacturing process, analytical method and stability models, was counted in T2 under rule set v0.4.2. It is process development, so under the v0.4.3 theme 2 boundary it is out of scope and not counted [>BMS-R23].

Outside the R&D count, a manager for GenAI and data science in pharmaceutical product development is an enterprise role under C42-21, because the duties name no discovery activity and no lab or scientific data system [>BMS-R22]. The newly created Drug Development AI Nexus lead and several clinical development and clinical operations AI roles are enterprise roles [>BMS-R15, >R19, >R35, >R40, >R42]. Translational computational biology and translational informatics roles are out of scope under C42-8 and C42-20 [>BMS-R28, >R29]. Plant roles include lab automation and QC automation in cell therapy manufacturing and a Veeva LIMS role for QC laboratories at the RayzeBio Indianapolis plant [>BMS-R34, >R16, >R20]. The internship is with a QC laboratory data systems team at the Cruiserath site [>BMS-R09].

Posting dates come from a field in the careers API data that many postings share with the snapshot date. It may record a refresh, so the dates are not reliable.

No LinkedIn company insights were collected for BMS.

Technology mentions in recruitment

Mentions are classified from the exact posting wording. They do not establish installation, an incumbent contract or company-wide use. The table records 199 BMS mentions: 16 explicit use, 12 integration or environment, 30 required experience, 2 preferred experience, 3 planned implementation, 132 alternatives and 4 unclear. The seven R&D theme postings carry 24 of them: 1 explicit use, 9 required experience, 1 preferred experience and 13 alternatives. - Named as an existing system: the PD Model Hub, within which the MLOps engineer is to deploy, monitor and retire models. The posting does not say whether it is in-house or vendor software [>BMS-R23]. This is role-specific wording. - Integration responsibilities: none recorded on the R&D postings. - Required experience: Databricks, AWS, MLflow, Docker, Kubernetes, GitHub Actions, Azure DevOps, Evidently AI and PyTorch, for the Model Hub role [>BMS-R23]. - Preferred experience: Model Context Protocol (MCP) auth patterns [>BMS-R25]. - Lists of examples: Hamilton VENUS, Tecan FluentControl or Agilent VWorks/Bravo for automation scripting; Green Button Go, Cellario, Momentum or Overlord as orchestration or scheduling software; "ELN, LIMS, or registration systems (e.g., Benchling)" [>BMS-R39]; Claude coding as an example of an AI agent [>BMS-R27].

No ELN or LIMS product is named as in use in the R&D postings, and no category (ELN, LIMS, cloud) is recorded as explicit use on them. Outside R&D, explicit-use mentions include Veeva LIMS for the RayzeBio plant QC laboratories [>BMS-R20], DeltaV and PI at Devens [>BMS-R33], Claude, Claude Code and Databricks in supply risk analytics [>BMS-R12], Claude in clinical operations GenAI work [>BMS-R40], and Xenium, Visium and Visium HD as spatial platforms a translational team "runs today" [>BMS-R28]. Two category mentions outside R&D are recorded as explicit use because the posting states the employer's own system: MES at Devens and AWS cloud in clinical engineering [>BMS-R08, >R42].

Deployment evidence

• BMS's July 2026 release says the expansion "builds on nearly three years of collaboration that began when BMS first deployed NVIDIA DGX SuperPOD infrastructure" for R&D. The start date is inferred from that wording, not stated, and falls before both windows [>BMS-E01].
• The same release says AI agents automate target identification and validation, and that through Predict First "AI informs the design of every small molecule program" [>BMS-E07]. Earlier BMS stories say BMS predicts before it synthesises "the majority of the molecules" (February 2025) and that all its small-molecule degrader and LDD programmes are supported by Predict First (January 2026) [>BMS-E05, >E06].
• An NVIDIA case study dated 11 September 2026 says "BMS established an AI center of excellence and an AI factory powered by NVIDIA DGX SuperPOD", that Equinix managed the infrastructure and colocation, and reports "55% overall cost savings" against BMS's prior computing model [>BMS-E03]. This is a vendor source quoting BMS staff, and it gives no creation date. Equinix's Q3 2025 results release names BMS as a customer that runs AI workloads on its infrastructure, with no BMS quote [>BMS-E04].
• A March 2025 BMS story says IPS uses large language models "to automate what we call target reports" [>BMS-E08].
• Apheris reported on 14 September 2026 that BMS and four other companies "each fine-tuned OpenFold3 Preview2 locally on their own structures" in the federated initiative [>BMS-E13]. The report says the participating companies funded the consortium. The date the work was done is not stated.
• insitro reported in March 2026 that BMS nominated two more ALS targets identified through insitro's Virtual Human platform, with a US$10 million milestone paid to insitro, after a first target nominated in December 2024 [>BMS-E23, >E22].
• Evotec reported payments under the protein degradation partnership: US$75 million in April 2025, a US$5 million milestone after an IND acceptance in November 2025 and a US$10 million milestone for a Phase 1 start in March 2026 [>BMS-E16, >E17, >E18]. These are partner releases with no BMS quote.
• The 2025 Form 10-K records that BMS completed the Orbital acquisition in December 2025 [>BMS-E20].
• T2: a July 2025 BMS story shows a robotic machine that "performs automated cell culture, growing stem cells" studied to discover targets for neurological diseases [>BMS-E26]. A Rescale release of October 2025 quotes a BMS Research Fellow on Rescale modelling workflows used in process development; this is process development, recorded as context under the v0.4.3 theme 2 boundary, and is not counted [>BMS-E27].
• An undated BMS locations page describes the CITRE site in Seville as housing computational research teams and "dedicated computational research infrastructure" [>BMS-E10].
• Apart from the vendor cost figure and BMS's own statements of scope, we found no adoption or benefit metric for any of these items in the reviewed sources.

Context

Sally John joined BMS as senior vice president in 2024 to lead the Information and Predictive Sciences team; no appointment date is stated [>BMS-E09]. The NVIDIA release quotes Greg Meyers as Chief Digital and Technology Officer and Robert Plenge as Executive Vice President and Chief Research Officer [>BMS-E02]. Payal Sheth is quoted as Vice President, Discovery Biotherapeutics and Lead Discovery and Optimization in October 2025 and as Senior Vice President, Therapeutic Discovery Sciences in September 2026; we found no appointment date [>BMS-E12, >E14]. Stephen Johnson is quoted as Vice President, Computational Sciences [>BMS-E24]. The NVIDIA case study quotes BMS's Director of Research Computing [>BMS-E03].

Under C42-9 the Anthropic agreement is context, not a T1 commitment. On 20 May 2026 BMS announced an agreement to deploy Claude across its research, clinical development, manufacturing, commercial and corporate functions, for more than 30,000 employees [>BMS-E28]. Research use for target identification sits under a statement that BMS "will evaluate the potential" for Claude to serve as an agentic layer, so discovery scope is a use to be evaluated, not a stated scope of the agreement [>BMS-E28].

Other enterprise statements: the 2025 Form 10-K says BMS is "infusing artificial intelligence throughout our R&D process" [>BMS-E34]. The 2025 annual report says AI within R&D puts BMS on track to reduce clinical development timelines by about 30% [>BMS-E36]. The JPM 2026 deck lists "Deploying AI enterprise-wide" among priorities, and the Q2 2026 deck links AI execution to the Anthropic and NVIDIA partnerships [>BMS-E37, >E38]. Earlier statements cover custom ChatGPT and GenAI tools for staff, AI use across the organisation and workforce upskilling [>BMS-E30, >E31, >E32, >E33, >E35]. Clinical, manufacturing and commercial AI items are context: radiology AI with Microsoft for early lung cancer detection, more than 30 digital and AI use cases at the Devens manufacturing site, and Veeva AI agents in commercial engagement [>BMS-E29, >E39, >E42]. A posting first seen on 6 October 2026 describes the Drug Development AI Nexus decision integration lead as "a newly created leadership role"; its remit is drug development decisions, so it is context [>BMS-E41, >R15].

Under AR7a, BMS's sponsorship of the UK Biobank Pharma Proteomics Project, as one of 14 companies, is data access and context [>BMS-E40]. The immunology company created with Bain Capital in July 2025 is a venture with no AI platform stated (C42-3) [>BMS-E43]. Two Evotec neuroscience payments are excluded because no AI or computational discovery work is stated [>BMS-E44, >E45].

BioProcess International reported in August 2026 that a $380 million partnership between BMS and Cellares "will be terminated", and that BMS cited the manufacturing requirements of its CAR T-cell therapy Breyanzi. This is GMP manufacturing automation, outside T2 and not linked to either theme [>BMS-E46]. A Korean media report of 4 November 2025 says BMS stopped developing a PKC-theta inhibitor discovered through its Exscientia AI partnership; under C42-2 this is context, an AI-derived asset [>BMS-E47]. We identified no qualifying pause or cancellation in the two themes in the searched sources. Both reviewers ran the negative-event search for each theme. That is a search result, not evidence that none occurred.

Coverage limits

• news.bms.com, including its RSS feed, returned a Cloudflare challenge to both reviewers. BMS releases were read from the IR feed on bristolmyers2016ir.q4web.com, whose HTML detail pages were also blocked; the feed's JSON endpoints returned full release bodies. The feed omits deal releases that the 2025 10-K records (2seventy bio, Philochem, BioArctic), so the press release archive family is partial.
• Call transcripts were not retrieved, so the results presentations and transcripts family is partial. Only decks and filings were read.
• Business Wire blocked access, even to robots.txt, and GlobeNewswire timed out. The PR Newswire search was capped at its 400 newest results, which did not reach W0. Partner releases were read on Apheris, Evotec, insitro and Chai Discovery sites and in trade press. The OpenFold date comes from a Business Wire item ID in a link, and the release text was not read [>BMS-E11]. Schrödinger's investor site blocked access, so the agreement date is unknown [>BMS-E24]. Fierce Biotech, Pharmaceutical Technology, European Pharmaceutical Review and healthexec.com also blocked access.
• SEC EDGAR blocked both reviewers. 10-K PDFs on bms.com were read, and their first-public dates rest on PDF creation dates, so they are uncertain [>BMS-E20, >E31, >E34, >E35].
• Search budget: each reviewer ran at most nine web searches in phase 2. Complete families: annual report and 10-K, consortium, academic and negative-event searches; the per-event-type search is complete for T1. Partial families: company press release archive, results transcripts, partner newswires, corporate venture arm, and the per-event-type search for T2.
• A dedicated T2 search for laboratory automation, electronic lab notebooks, LIMS and scientific data platforms ran on 8 October 2026, after the evidence cutoff, with the press release list read in a browser. It found no further eligible BMS disclosure. The queries and pages read are recorded in Ortent's search log.
• Searches not run for lack of budget in phase 2: a search to date the Schrödinger agreement; a search for a readable copy of the OpenFold release; and searches for Business Wire copies of the NVIDIA and Chai releases and for GlobeNewswire and PR Newswire items in W0. The T2 counts rest on the IR feed, BMS stories, filings, decks and capacity searches that included only two T2 terms, and on the 8 October supplementary search.
• No BMS venture-arm site was identified, and investee newsrooms, university news pages and the IHI, SGC and Pistoia member pages were not fetched directly. The phase 2 web searches had no date filter, and their results skew to 2026.
• The CITRE page is undated, so its disclosure window is uncertain [>BMS-E10]. Rows drawn from job postings take 6 October 2026 as their first public date, because no page shows a date-posted field [>BMS-E41].
• Patents are not among the searched source families in the coverage table. Publication counts are raw Europe PMC query counts and are not events.
• Recruitment: both reviewers enumerated jobs.bms.com through its search and position-details endpoints; one apply endpoint was blocked. Only the 45 postings classified are in the table.
• Two independent reviewers searched BMS, and the final rows are their source-verified union. That union contains rows that each reviewer missed, but it does not show that both did not miss the same events. OpenFold was found by one reviewer, and the Evotec expansion, the insitro extension and Schrödinger by the other [>BMS-E11, >E15, >E21, >E24].

Two questions to investigate (Ortent's interpretation)

1. Lab data and lab automation roles sit in different units. A Molecular Invention Data Product Team is building discovery data products with FAIR metadata [>BMS-R17]. Discovery Biotherapeutics is recruiting to lead high-throughput biologics production with laboratory automation and closed-loop discovery workflows, naming ELN, LIMS and orchestration software only as examples [>BMS-R39]. Pharmaceutical Product Development runs a Model Hub for process and analytical models, which is process development outside T2 under the v0.4.3 boundary [>BMS-R23], and a BMS story shows automated stem-cell culture in neuroscience discovery [>BMS-E26]. What is the remit of each unit for lab data systems and lab orchestration software? Who decides on ELN, LIMS, registration and scheduling software, and how are the interfaces between these discovery data products, the lab automation teams and the compute that BMS says will train models on its proprietary data governed [>BMS-E02]?
2. BMS has external AI-discovery agreements with Chai Discovery, Schrödinger, insitro and Evotec, and joined the OpenFold and Apheris federated initiatives [>BMS-E25, >E24, >E21, >E15, >E11, >E12, >E14]. Internally, Computational Sciences and IPS are recruiting machine learning, computational design and agentic AI roles [>BMS-R02, >R26, >R25, >R27], and the NVIDIA compute expansion was announced with the Chief Digital and Technology Officer quoted [>BMS-E02]. The Machine Learning Sciences lead is to evaluate "external technologies" [>BMS-R02], and the enterprise Anthropic agreement includes an evaluation of research use [>BMS-E28]. Who holds decision rights over licensing an external AI discovery model or agentic tool rather than building on BMS's own models and compute, and do such agreements pass through Computational Sciences, IPS, the digital and technology organisation, business development or procurement? How is funding for these decisions set, given that no value was disclosed for the Vera Rubin expansion [>BMS-E02]?

Sources for the main claims

• BMS-E15 Evotec, partnership expanded beyond oncology: https://www.evotec.com/news/evotec-and-bristol-myers-squibb-expand-proteomics-partnership
• BMS-E11 OpenFold news page (eight new members): https://openfold.io/news.html
• BMS-E12 Apheris, Federated OpenFold3 Initiative expansion: https://www.apheris.com/resources/aisb-network-expands-federated-openfold3-initiative-with-three-new-pharma-contributors
• BMS-E19 BMS and Orbital Therapeutics, acquisition agreement (BMS IR feed): https://bristolmyers2016ir.q4web.com/iframes/press-releases/press-release-details/2025/Bristol-Myers-Squibb-Strengthens-and-Diversifies-Cell-Therapy-Portfolio-with-Acquisition-of-Orbital-Therapeutics/default.aspx
• BMS-E21 insitro and BMS, collaboration extension (insitro newsroom copy): URL and archived copy in the BMS events table
• BMS-E02 BMS and NVIDIA, Vera Rubin expansion (BMS IR feed): https://bristolmyers2016ir.q4web.com/iframes/press-releases/press-release-details/2026/Bristol-Myers-Squibb-to-Build-the-Most-Powerful-AI-Factory-in-Life-Sciences-with-NVIDIA/default.aspx
• BMS-E24 Contract Pharma, Schrödinger Bunsen agreement: https://www.contractpharma.com/?p=91760
• BMS-E25 Chai Discovery collaboration (Chai newsroom): https://www.chaidiscovery.com/news/bms-partnership
• BMS-E14 Apheris, AISB Bind launch: https://www.apheris.com/resources/aisb-network-launches-aisb-bind-to-improve-prediction-of-how-molecules-bind-for-virtual-screening-and-lead-optimization
• BMS-E03 NVIDIA case study: https://www.nvidia.com/en-us/case-studies/computational-science-accelerates-research-innovation-at-bristol-myers-squibb/
• BMS-E13 Apheris, federated OpenFold3 results: https://www.apheris.com/resources/federated-training-dramatically-improves-the-accuracy-of-protein-ligand-co-folding-on-private-pharma-structures
• Other event rows and all requisitions: URLs and archived copies in the BMS tables. No SRC source files are cited for BMS.

Source index

Each ID in brackets links to the online source for that row. Job posting links come from a snapshot of the company job site on 6 October 2026; postings that have closed since are marked and no longer linked. Where a company site blocked access, the link goes to the copy that was read.