On 10 September 2026, OpenAI published details of Dr. César de la Fuente's laboratory using Codex and ChatGPT to search living and extinct genomes for antimicrobial candidates S¹. The work targets drug-resistant infections, which grow harder to treat as bacteria evolve resistance faster than new drugs reach the market. What the blog post does not say is whether any of these AI-found candidates have actually worked in a lab dish, let alone a human body.

My read: This is the first application I've seen where Codex, a tool built to write code in a terminal, is pointed at a biological problem rather than a software one. I don't buy the implicit promise that ChatGPT can "discover" antibiotics in any meaningful sense yet, because the evidence is a single corporate blog post with no peer-reviewed paper, no named molecule, and no clinical trial behind it. What I do find plausible is that these tools speed up the boring parts of genomic search, the pattern-matching and data wrangling that eats weeks of a researcher's time. The question worth watching is whether any candidate from this pipeline survives testing in a real lab.

A coding agent repurposed for biology

Codex is a lightweight coding agent that runs in a terminal P⁴. OpenAI revealed that a research lab is using that same tool, alongside ChatGPT, not to write software but to hunt for molecules that might kill bacteria S¹.

The de la Fuente lab examines the DNA of both living and extinct organisms to find sequences that could produce antimicrobial effects S¹. The reasoning is simple: nature has operated its own drug-discovery system for billions of years, and some of the most valuable antimicrobial compounds originated from organisms that are now extinct. AI systems capable of quickly scanning and identifying patterns in massive genomic datasets might reduce the search time from years to days.

Another OpenAI post from 9 September notes the team is "using new tools in the fight against antimicrobial resistance" P³. The wording is deliberate, highlighting tools rather than cures.

Why antibiotic discovery needs the help

Conventional antibiotic development is slow and offers poor commercial returns, resulting in fewer companies pursuing it. Any system that accelerates the initial phase of identifying promising molecules for testing is significant, as that step consumes the majority of time and funding.

The de la Fuente lab's method involves using AI for the initial screening: analyzing genomes, identifying sequences that appear to generate antimicrobial compounds, and passing those candidates to human scientists for laboratory evaluation S¹. The AI does not determine effectiveness; it simply reduces the scope of the search.

What to do about it

For a microbiology lab already tackling antimicrobial resistance, the lesson is not to simply install ChatGPT and expect a drug. Rather, general-purpose AI systems can be repurposed for biological data challenges with minimal custom development. A laboratory conducting genomic searches might evaluate whether a coding agent such as Codex, which manages structured data and pattern recognition in a terminal environment, can accelerate their existing screening process.

Imagine a university microbiology team that currently dedicates weeks to executing BLAST searches, the standard method for comparing genetic sequences, across hundreds of genomes. If an AI agent can automate the scripting, handle the batch processing, and highlight the most promising results for human inspection, the researchers can reach the lab-bench testing phase more quickly. That represents the true benefit: condensing the tedious intermediate steps, not substituting the researcher.

The actionable step this week: review the OpenAI post S¹, then assess whether your laboratory's current genomic search workflow contains repetitive, scriptable steps that are constrained by human availability. Those are the duties an agent like Codex is designed to handle.

What we don't know yet

The evidence is severely limited. The assertions originate from a single OpenAI corporate blog post S¹ and a nearly identical companion article P³. No peer-reviewed publication supports the announcement. No particular antimicrobial molecule is identified. No laboratory test results are referenced. No clinical trials are noted.

The difference between "using AI to search" and "AI discovered something useful" represents the entire gap. OpenAI's post illustrates the former. The latter remains unverified.

The next indicator: whether de la Fuente's lab releases peer-reviewed findings from this pipeline in the upcoming months. We will compare any future paper against the assertions in this post. Subscribe to receive that follow-up when it is published.

Sources: S1 — How a researcher uses Codex and ChatGPT to search for new antimicrobia · S2 — How a researcher uses Codex and ChatGPT to search for new antimicrobia · P3 — Accelerating antibiotic discovery with ChatGPT · P4 — openai/codex · P5 — openai/gpt-oss · P6 — ArtmeScienceLab/FonTS

More from Not A Tech Guy

Generated from an audited evidence pack with primary-source research. Social-media items are discussion signals, not verified facts. Nothing here is financial, legal or medical advice.