GPT-5 in a Lab Coat: OpenAI Goes After Big Science

In October, OpenAI officially announced a program to bring AI into scientific research, starting with a partnership with Los Alamos National Laboratory — the same lab where the US nuclear shield was built. This is not a marketing stunt. Sam Altman's company is shifting its focus from office work to people in lab coats, entering a field where a mistake costs far more than a failed social media post.

From office assistant to the lab

In the three years since ChatGPT debuted, users have grown used to handing off routine work to it, from emails to code. For a long time, the conventional wisdom held that language models were strong at humanities tasks but struggled with the rigorous logic of the natural sciences. After the success of o1, a model that learned to reason, OpenAI became convinced that the potential for deep research is enormous.

In October, that conviction turned into a concrete step: the official launch of a program to bring AI into scientific research and a partnership with Los Alamos National Laboratory. The choice of partner is symbolic. This is where the US nuclear shield was once forged, so the bar for accuracy and reliability has historically been higher there than in any office task.

What exactly OpenAI is doing at Los Alamos

This is not about a chatbot for finding papers. OpenAI is building specialized benchmarks and tools to evaluate how effectively models can assist with biological and chemical experiments. That includes analyzing complex molecular structures, predicting reaction outcomes, and helping plan lab tests. Crucially, the company is not just opening up API access. It is building an entire ecosystem in which AI becomes a full participant in the research process rather than an external tool for quick information lookup.

The race with Google DeepMind

The timing is no accident: competition is the driving force here. With AlphaFold, Google DeepMind has already revolutionized biology by predicting the structures of nearly all known proteins. OpenAI cannot afford to let a rival own the title of the most scientific AI by itself. If ChatGPT becomes the standard in labs the way it became the standard in offices, the company will gain access to the most valuable resource of all: data on real physical processes. That is critical for building AGI, which must understand the world not only through text but also through the laws of matter.

Biosecurity risks

Using AI in biology and chemistry inevitably raises biosecurity questions: a model's ability to design a new drug could, in theory, also be used to create something far less humane. That is exactly why OpenAI puts so much emphasis on its partnerships with government bodies like Los Alamos — it is a way to show regulators that the process is under control. But behind the safety facade lies a fierce fight for intellectual leadership in the industry. If the company's next model helps achieve a real breakthrough in medicine or energy, the question of AI's practical usefulness will be settled for good.

When did

OpenAI launch its partnership program with Los Alamos?

In October. The company officially announced a program to bring AI into scientific research, starting a partnership with Los Alamos National Laboratory, where the US nuclear shield was once built.

How does

OpenAI's approach to science differ from Google DeepMind's?

DeepMind, with AlphaFold, focused on protein structure prediction and has already predicted nearly all known protein structures. OpenAI is going broader: it is developing benchmarks and tools to assist with biological and chemical experiments — analyzing molecular structures, predicting reaction outcomes, and planning lab tests — and embedding AI directly into the research process, not just the prediction stage.

What risks does using AI in biology and chemistry carry?

The main risk is biosecurity: a model's ability to design a new drug could, in theory, be used to create less humane substances. That is why OpenAI is building partnerships with government bodies such as Los Alamos — a signal to regulators that the technology's development is under control.