Method

What counts as an entry

• A named pharma company (or its partner) described an AI initiative in public.
• The description comes from a primary source — a press release, earnings call, annual report, investor presentation, or peer-reviewed paper. Trade press is used when it directly quotes the company.
• A LinkedIn post by an employee counts only if it is clearly on the record and the company is named.
• One initiative per entry. Multi-initiative press releases are split.
• No source, no entry. No estimate, no prediction.

How we write entries

• The initiative sentence is our own words. The evidence is always the company's words, quoted directly.
• We never upgrade status. A pilot stays a pilot until a later source says otherwise.
• Dates are disclosure dates, not launch dates, unless the source gives a launch date.
• Vendor-published figures are labelled "vendor claim." Secondhand reports are labelled "Secondhand."

Sources and archiving

• Every entry links to its source. We prefer company-issued sources; we use trade press only when it quotes the company directly.
• We save an archived copy (Wayback Machine or archive.today) on the day we add the entry.
• If a source link goes dead, we show the archived copy and note it.

Update cadence

Updated monthly. We re-verify every source link at least once a quarter and update the "Verified on" date on each entry when we do.

Corrections

Corrections are public. If something is wrong, we fix it, date the correction, and log it in the changelog.

Spot an error or a missing source? Tell me and I'll fix it quickly.

Solution areas

Each entry is assigned to one of eleven solution areas. The assignment uses the company's own framing where possible.

1. AI drug discovery. A company's own AI work for target discovery, molecule design and other drug discovery research. External research collaborations stay under Research partnerships.
2. Clinical development
3. Diagnostics and clinical decision support. Imaging, diagnostic devices, digital biomarkers, and tools that support a clinical or pre-diagnosis decision.
4. Regulatory, medical writing and pharmacovigilance
5. Manufacturing and quality
6. Supply chain and distribution
7. Commercial and medical affairs
8. Enterprise, IT and knowledge work
9. AI infrastructure. Compute, platforms, and tooling for building or running AI when the entry does not name a separate business use.
10. AI governance and enablement. Councils, rules, and programs that govern how AI is used or build people's ability to use it.
11. Other

How to read the leaderboard

The counts and rankings are directional indicators, not definitive measurements. They measure public disclosure volume only, not how much AI a company uses or how good it is.

Disclosure is selective. Companies choose what to publish, and many have competitive, confidentiality or regulatory reasons not to describe their AI work in detail.

Counts also depend on how each company publishes. Some announce through press releases, others through conference talks, annual reports or investor presentations, and the Tracker only counts disclosures that are public and sourced. One large program can appear as a single entry, while a company that announces often, or announces each phase separately, can collect many entries for similar work.

A company with fewer disclosures may have more AI activity that isn't publicly documented. Read the leaderboard as a view of disclosure trends over time, not a precise measure of any company's AI capability.

Independence

An independent register of public disclosures. Not affiliated with any regulator, government AI register, or the companies listed.

Company names are used only to identify the source of each public disclosure. Inclusion does not imply endorsement, partnership, or any relationship with the author.