How Codex and ChatGPT Are Helping Researchers Search for New Antimicrobial Molecules

Drug-resistant infections remain a major scientific and public health challenge. Researchers are now turning to artificial intelligence to explore one of biology's largest search spaces: the genomes and proteins of living and extinct organisms.

In a September 10, 2026 article, OpenAI highlighted work by bioengineer César de la Fuente and his lab, which uses AI to search biological sequences for molecules that could potentially become new antimicrobials. The research combines specialised deep-learning models with ChatGPT and Codex to support tasks ranging from hypothesis development and coding to dataset processing and analysis.

The approach does not replace laboratory science. Instead, AI is being used to help researchers identify promising candidates from enormous biological datasets before those candidates undergo experimental testing.

Searching biology as an information system

The research starts from the idea that biological information can be analysed through the sequences that encode proteins and peptides.

According to OpenAI, de la Fuente's lab trains deep-learning models to recognise patterns in biological sequences. These models can search large genome and protein datasets for potential antimicrobial candidates.

OpenAI says this approach can reduce the initial search for candidate molecules from years to hours. However, this refers to the early computational search stage rather than the complete process of developing an approved medicine.

That distinction is important.

Finding a promising sequence is only an early step. Researchers still need to determine whether a candidate actually kills the target microorganism, how much is required, how it interacts with human cells and whether microbes can develop resistance to it.

Candidates may also require chemical optimisation, toxicity testing, stability assessment, manufacturing development, regulatory review and clinical trials before they could become approved antimicrobial drugs.

AI tackles the biological "needle in a haystack"

A major challenge in antimicrobial discovery is the enormous number of possible biological sequences.

Scientists have historically searched for antimicrobial compounds in sources including plants, animals, microbes, insects, water and soil. Modern digital databases allow researchers to examine biological information across a much broader portion of the tree of life.

The problem is no longer simply obtaining information. It is identifying useful signals within that information.

AI can help with this type of search by scanning large datasets, identifying patterns and prioritising candidates for experimental investigation. This gives researchers a way to narrow an enormous computational search space before committing laboratory resources.

But AI-generated predictions are not automatically biological discoveries.

OpenAI's account of the research stresses that experimental evidence remains essential. A candidate predicted by an AI system must ultimately be tested in the physical world to determine whether the prediction holds.

ChatGPT and Codex become research collaborators

The lab is not using AI only for biological prediction.

According to OpenAI, researchers use ChatGPT and Codex to brainstorm hypotheses, write and refine code, process datasets, analyse results and connect concepts across scientific disciplines.

This is particularly relevant to a research team that combines expertise in biology, chemistry, computer science and engineering.

Different researchers may have deep knowledge in one field but less experience in another. OpenAI says ChatGPT and Codex can help bridge these gaps by assisting researchers with unfamiliar terminology, methods and technical workflows.

The tools can also help with downloading, organising and preprocessing large genome datasets.

OpenAI says researchers use ChatGPT in their native languages as well, which can lower communication barriers within scientific workflows.

A collaborative sounding board, not a replacement for scientists

De la Fuente describes ChatGPT as a brainstorming partner that can help shape hypotheses.

The lab's researchers can feed ideas into their shared ChatGPT workspace, including ideas that may ultimately prove unsuccessful. OpenAI describes the workspace as a collaborative sounding board for people approaching research problems from different perspectives.

At the same time, the researcher emphasises the need to check AI-generated information for accuracy.

That limitation is particularly significant in life sciences. A plausible computational result is not enough to establish that a molecule works safely or effectively in a biological system.

The research therefore illustrates a hybrid model: AI helps researchers search, reason, code and organise information, while experiments provide the evidence needed to validate those predictions.

Where AI discovery ends and laboratory science begins

The antimicrobial discovery process described by OpenAI can be understood as a series of increasingly demanding validation stages.

First, computational systems search biological sequences for potentially interesting candidates.

Next, researchers need to establish whether those candidates demonstrate antimicrobial activity.

Further investigation can examine effectiveness, toxicity, resistance, stability and other properties. Chemical optimisation and manufacturing considerations may also become important.

Even after these stages, candidates must pass regulatory and clinical processes before they can become approved medicines.

This means AI can accelerate parts of discovery without eliminating the experimental and regulatory work required to develop a drug.

A broader shift toward AI-assisted scientific discovery

The work highlighted by OpenAI reflects a broader shift toward using AI not only to generate answers but also to help researchers navigate increasingly large scientific search spaces.

For enterprises and research organisations, this could mean that the value of AI increasingly depends on how well it fits into existing expert workflows.

In this example, the important combination is not simply "AI discovers drugs." It is the interaction between computational models, general-purpose AI assistants, domain specialists and laboratory experiments.

The research also demonstrates why human expertise remains central. AI can help identify patterns and accelerate information-heavy tasks, but experimental evidence determines whether a computational prediction corresponds to a real biological effect.

De la Fuente compares this evolution with earlier scientific tools such as telescopes and microscopes, which expanded humanity's ability to observe and understand the world. OpenAI presents AI-assisted biology as another stage in that longer history of scientific tool development.

What this means for AI and life sciences

The announcement highlights a broader industry shift toward AI-assisted scientific discovery.

The immediate opportunity is not necessarily to automate the entire drug-development process. Instead, AI can help researchers explore larger datasets, formulate hypotheses, automate parts of technical workflows and prioritise candidates for further investigation.

For life-science organisations, this could make the ability to combine AI expertise with domain knowledge increasingly important.

The research also reinforces a key principle for high-stakes AI applications: computational predictions need to be evaluated against real-world evidence.

AI can help researchers search biology faster, but the laboratory remains essential for determining whether those discoveries actually work.