Where flagged GenAI disclosures appear first
As of Oct 2026, flagged GenAI entries in the Pharma AI Tracker appear most often in enterprise tools, and enterprise is often the earliest area.
In the public disclosures recorded in the Pharma AI Tracker, flagged GenAI entries appear most often in enterprise tools. For a plurality of companies with flagged entries, enterprise is also the earliest flagged area; this is disclosure timing, not adoption timing. As of Oct 2026, the keyword rule flags 118 of 349 approved entries, or 33.8%, across 35 of 49 companies.
What stands out
As of Oct 2026, Enterprise, IT and knowledge work has 36 flagged entries from 20 companies and is the single earliest flagged area for 12 of 35 companies with flagged entries. GSK and Johnson & Johnson tie across areas and are not assigned to an area. Next are Research partnerships for 5 companies, AI drug discovery for 4, and Clinical development for 4.
Three of those 12, Celltrion, Roche and Viatris, have flagged rows that match only on chatbot or agent words: Celltrion and Viatris on chatbot, and Roche on agentic.
Amgen expanded Microsoft 365 Copilot across its workforce in May 2024, from a company source, in production and labelled non-GxP. Novo Nordisk listed internal ChatGPT and other productivity tools in Mar 2024, from a company source, in production and labelled non-GxP. Takeda described myAibou as an internal assistant for routine knowledge work in Jul 2024, from a company source, in production and labelled non-GxP. Pfizer’s Vox was presented as an internal base for multiple uses in Nov 2023, from a vendor claim, in production with GxP unclear. Sanofi said Concierge gives employees access to more than 20,000 data points and saves users an average of two hours a week in Feb 2026, from a company source, in production with GxP unclear.
As of Oct 2026, removing entries matched only by assistant, chatbot, agentic, AI agent, multi-agent or Agentforce leaves 80 entries across 30 companies. Celltrion, Eli Lilly, Roche, Viatris and argenx drop out. Enterprise still leads with 28 entries and is earliest for 9 companies, down from 12.
This is a public record, not an adoption curve. The Tracker is concentrated in recent disclosures: 254 of 349 entries (72.8%) are from 2025 or 2026, and 2026 is partial through September 29.
Public GxP-labelled disclosures are quieter
As of Oct 2026, of 118 flagged entries, 6 carry a GxP label, 46 say GxP unclear, 63 are non-GxP, and 3 are mixed. Regulatory, medical writing and pharmacovigilance has 10, while Manufacturing and quality has 8. The 6 GxP-labelled entries split between Manufacturing and quality, with 4, and Regulatory, medical writing and pharmacovigilance, with 2. All 9 in Clinical development say GxP unclear.
Sanofi’s PQR writing tool helps write annual product quality reviews in Jun 2025, from a company source, in production and labelled GxP. Merck & Co. tested generative AI to support manufacturing deviation work with human oversight in Nov 2025, from a vendor claim, in a pilot and labelled GxP. Shionogi tested a generative writing aid for regulatory documents with Hitachi in Feb 2026, from a company source, in a pilot and labelled GxP.
What a validator should take from this
The Sanofi pair makes the practical point. The PQR tool is labelled GxP in Jun 2025, while Concierge is GxP unclear in Feb 2026. Both rows say generative. Ask which label applies before deciding what validation a tool needs.
The note on the routing layer covers model choice, logging, and controlled use. The leaderboard gives the public record across all AI entries.
Here, “first” means the earliest date named in the Tracker, not a forensic chronology. The Tracker records public company disclosures, vendor claims and secondhand reports; it does not measure internal adoption. A missing or thin row is not evidence of absence. As of Oct 2026, 14 of 49 companies have no flagged entry. Vendor claims and secondhand rows are labelled.
How I counted
The Tracker has no GenAI field. I searched only the initiative and summary text of each approved entry using case-insensitive, whole-word regular expressions for: generative; GenAI; gen AI, also gen-AI; LLM and LLMs; large language model and large language models; language model and language models; copilot and copilots, also co-pilot; ChatGPT; GPT, also versions such as GPT-4; Claude; Gemini; Bedrock; OpenAI; Anthropic; assistant and assistants; agentic; AI agent and AI agents, also AI-agent; multi-agent and multi-agents; chatbot and chatbots; conversational AI; text-to-SQL; Agentforce. There are no hand exclusions. It is a crude keyword filter, so a few matches, including one Takeda equipment monitoring row from Dec 2025, are not generative AI, and some generative work will be missed. Counts are of approved entries as of Oct 2026, using the Tracker’s method.
Related reading: pharma is hiring people to steer AI agents.
Want to find where GenAI fits in your company first? That is the work I do.