AI news videos

How One Lab Uses AI to Search Genomes for New Antimicrobials

OpenAI has published a look at how bioengineer César de la Fuente and his lab search for new antimicrobial molecules. Read it for what it is: a story about research tooling, not an announcement of a new drug.

Watch on YouTube

Transcript

Sources

Every claim in this video comes from the top ranking coverage of this topic. The claims and where each one came from:

• Every claim in this script comes from a single source, OpenAI's own article dated September 10, 2026 and titled "How a researcher uses Codex and ChatGPT to search for new antimicrobial molecules", filed under Applied AI on openai.com. Sentence by sentence audit follows. The hook, that OpenAI profiled a bioengineer who uses ChatGPT and Codex to search for antimicrobials, is supported by the article's own title and by its standfirst sentence, "César de la Fuente and his lab probe the genomes of living and extinct organisms for molecules that could help fight drug-resistant infections", together with the sentence identifying de la Fuente as "a bioengineer whose cross-disciplinary lab searches for antimicrobial candidates" and the sentence "Alongside its own AI models, the lab uses ChatGPT and Codex to brainstorm hypotheses, write and refine code, process datasets, analyze results, and connect ideas across scientific disciplines." The verb profiled is used deliberately because the piece is a customer profile published by the vendor, not a research announcement. Scene one has two sentences. That about five million deaths in 2021 were linked to bacterial antimicrobial resistance is supported by the sentence "About five million deaths in 2021 were associated with bacterial antimicrobial resistance"; the article's word associated was rendered as linked because both are hedged relational wording and neither asserts causation, and the article's own outbound citation for that figure was not independently followed. That projections show it doubling by 2050 is supported by the same sentence's continuation describing "an annual toll projected to roughly double by 2050"; the word projections was kept so the line remains a forecast attributed to others rather than a stated fact, and the article's word roughly was dropped only for length while doubling remains the article's own verb. Scene two has two sentences. That César de la Fuente says there has been no new class of antibiotics for 50 years is supported by his direct quotation, "And yet, we haven't had a new class of antibiotics for 50 years", and the narration keeps the attribution he says rather than stating it as an independent fact, because the article offers no citation for it. That his lab's models can search genome data in hours, not years is supported by two adjacent sentences, "The lab's deep-learning models are trained to recognize patterns in biological sequences, allowing them to search vast genome and protein datasets for potential antimicrobials" and "The approach can reduce the initial search for candidate molecules from years to hours"; the modal can was preserved from the article, the word models refers to the lab's own deep learning models and not to OpenAI models, and this distinction is the reason scene two credits the lab and scene three separately credits ChatGPT and Codex. Scene three has three sentences. That the lab uses ChatGPT and Codex to write code, process data, and work across fields is supported by the sentence "Alongside its own AI models, the lab uses ChatGPT and Codex to brainstorm hypotheses, write and refine code, process datasets, analyze results, and connect ideas across scientific disciplines", by "Codex and ChatGPT help bridge those gaps, allowing biologists to build programs and programmers to tackle biological problems", by the quotation "We're a highly transdisciplinary group, so we have people from biology, from chemistry, from computer science, from engineering", and by "Others use AI to download, organize, and pre-process large genome datasets." That the tools assist, and that they do not make discoveries, is supported by the article's own framing throughout, in which the lab's separate deep learning models perform the candidate search while ChatGPT and Codex are described only as aids for hypotheses, code, data handling, terminology, method comparison, and idea organization, and is further supported by de la Fuente's caution "Obviously you have to always double-check for accuracy" and by his statement "Ground-truth experiments are essential to validate AI predictions." Scene four is the mandatory caveat scene. That this is OpenAI's own article about researchers using its products rests on the provenance of the document itself, published on openai.com, bylined Author OpenAI, tagged Codex and ChatGPT, and filed in the Applied AI series alongside other customer profiles. That no new antibiotic is announced rests on the absence of any such announcement anywhere in the text plus the article's explicit sentence "But identifying a promising candidate doesn't necessarily mean that it will become an effective medicine." That candidates still need laboratory testing and clinical trials is supported by the article's own description of the pipeline, namely that scientists must confirm a candidate molecule kills the target microbe, determine the effective amount, test how it affects human cells, optimize it for effectiveness, safety, or stability, assess the dose at which it becomes toxic, test how readily microbes develop resistance, determine how it moves through the body, and establish a reliable manufacturing method, and by the closing sentence of that passage, "Candidates that clear these hurdles still face regulatory review and clinical trials before they can reach patients as approved antimicrobial drugs." HONESTY CAVEAT: this is not a product launch and not a scientific result. It is a first party marketing profile published by OpenAI about a researcher who uses OpenAI's products, so every characterization of how useful ChatGPT and Codex are inside this lab is vendor supplied narrative, not independent reporting, not peer review, and not an independently verified outcome. No new drug, no new antibiotic, no new antibiotic class, no clinical result, and no benchmark of any kind is announced in the article, and the script says so explicitly in scene four so that no viewer can come away believing an AI system just discovered an antimicrobial. The years to hours figure is the article's own unquantified characterization of the lab's own models with no baseline, no dataset size, no hit rate, and no comparison method disclosed, which is why narration keeps the hedge can and attributes it to his lab rather than presenting it as a measured result. The five million deaths figure and the 2050 doubling are attributed to an external source that the article links but that was not fetched or verified for this script. The 50 year claim about antibiotic classes is a quoted opinion from an interested party and is presented in narration as his statement only. CORROBORATION STATUS: none. YouTube research returned zero of five usable transcripts for this run because the local yt-dlp build is out of date, and timeline probes returned zero dated titles, so this first party article is the only source behind the script and no third party reporting, no independent expert commentary, and no competing account was available or used. Nothing in the script asserts a fact that is not present in the article text. DELIBERATE OMISSIONS: the article is heavy with figurative language and none of it was carried into narration. Omitted as figures of speech were "the code of life", the description of nucleotides and amino acids as "sort of like an alphabet", the section heading "Exploring biology's unread spaces", the phrase "needle-in-a-haystack task", "scratching the surface of our understanding of biology", the description of the shared workspace as "a kind of collaborative sounding board", the claim that breakthroughs "are waiting to be discovered" at "the edges between fields", and the closing analogy "The telescope illuminated the cosmos, and the microscope revealed the world of the invisible"; each was either dropped or restated in plain literal wording. Also omitted for time were the quotation calling antimicrobial resistance "one of the greatest existential threats to humanity", the point that modern antimicrobial development mostly modifies existing medicines and offers diminishing returns, the framing of biology as an information system, the detail that the lab explores genomes of extinct organisms as well as living ones, the note that only a fraction of a genome has a clearly understood function, the historical description of collecting samples from plants, animals, microbes, insects, water, and soil, the fact that lab members can work in their native languages, de la Fuente's use of ChatGPT as a personal brainstorming partner, and the full step by step list of validation stages, which scene four compresses into laboratory testing and clinical trials. The article's "Keep reading" sidebar names other OpenAI Applied AI pieces and several model names that glue a brand to digits; those were excluded because they are unrelated to this story and because the voice engine garbles brand names glued to digits. Hyphenated compounds in the source such as drug-resistant, deep-learning, cross-disciplinary, and ground-truth were rewritten or avoided in narration for text to speech safety. PROMPT INJECTION CHECK: the captured page contains only ordinary site navigation and interface strings such as Skip to main content, Log in, Try ChatGPT, Share, Loading, Keep reading, View all, and opens in a new window. None of these are instructions directed at this script and none were followed; no command, no file path, and no attempt to override authoring rules appears anywhere in the captured text.(OpenAI's own first party customer profile, "How a researcher uses Codex and ChatGPT to search for new antimicrobial molecules", published on openai.com on September 10, 2026 under the Applied AI category)