Genmab puts ChatGPT Enterprise into 2,000 hands

Most corporate AI rollouts fizzle out because employees simply do not know what to do with a generic chat window. Biotech firm Genmab took a radically practical approach to avoid that trap.

After testing 1,000 ChatGPT Enterprise seats for less than a year, the antibody therapeutics maker just launched its "AI Everywhere" initiative, expanding the deployment to over 2,000 licenses—covering virtually everyone at the company. To get folks started, Genmab gave employees a dead-simple onboarding hack: paste in your job description and ask the model, "What can you do to help me?"

Over 100 custom GPTs tackle clinical reports and bio-translation

The company-wide push quickly translated into real work. Genmab teams have now built more than 100 custom GPTs spanning scientific literature summarization, routine document drafting, and data analytics.

According to an internal survey, employees save an average of 3.5 hours per week, logging roughly 120 conversations per user weekly. For a company that recently tripled in size and deals with staggering amounts of biological data, that is serious usage.

More importantly, they are pushing the model into regulated, high-stakes biotech tasks. Teams are using GPT-4o to draft clinical study reports (CSRs), patient safety narratives, and investigational new drug (IND) applications. By pairing the model directly with clinical trial data and expert feedback, the system iterates on its own drafts much like an apprentice scientist.

They are also leaning heavily on vision capabilities. GPT-4o now interprets documents containing mixed diagrams, numbers, and dense text to verify factual citations across medical publications—a grueling manual task that used to eat up weeks.

Another internal tool, dubbed AI Translator GPT, mimics a human translation desk. It splits the pipeline into distinct agents—a primary translator, an editor, and a biology subject-matter specialist. It even runs automated back-translation to spot hallucinated terms and routes English-to-Chinese translations through Japanese as a staging language, cutting turnaround times from weeks to minutes.

Cutting out the cloud middleman

Genmab also made an intentional choice on infrastructure. Instead of accessing models through a third-party cloud partner, they contracted directly with OpenAI.

Management noted that having a direct line to OpenAI's applied research and deployment teams helps them push frontier capabilities faster. Ahead of the expansion, Genmab vetted data privacy and compliance guardrails, setting up recurring reviews to keep clinical data locked down.

Why it matters

Biopharma is arguably the last place you would expect free-range LLM adoption. Regulatory filings and patient data carry zero margin for error. Genmab proves that with structured multi-agent workflows—like role-playing editors and back-translation checks—frontier models can handle heavily audited clinical paperwork without blowing up.

For enterprise leaders, Genmab's biggest lesson is adoption design. Handing employees a license with zero context rarely works; telling them to calibrate the AI against their exact job description created immediate buy-in. Spinning up 100 targeted GPTs also removed the friction of prompt engineering for non-technical scientists.

The real risk down the line is platform lock-in. Sinking your mission-critical clinical documentation deep into OpenAI’s proprietary stack makes switching providers nearly impossible. But when an enterprise genuinely recovers 3.5 hours per employee every single week, that trade-off is one most executives will happily make right now.