In In August 2024, Bin Xia Yang, portfolio director of pharmaceutical company Axcynsis Therapeutics, prepared to move the company's new tumor-optimized antibody drug conjugation into clinical trials. However, before the human exam began, one important step remained.
The Food and Drug Administration (FDA) requires drug developers to provide evidence that the drug is relatively risk-free before administering it to people. The form in question is called the New Drug (IND) application in clinical trials, and drug sponsors (usually manufacturers or marketers) must fill in hundreds to thousands of pages of material for the FDA to test. The document contains details about several aspects of the drug. Data from animal experiments, or from foreign human trials that provide safety and efficacy, an overview of how clinical trials are conducted, and details of how manufacturers produce the drug (if ultimately approved). “There's a lot of information. There's a lot of research reports,” Yang said.
IND applications are a tough need, but often burden drug developers. Pharmaceutical companies must file patents early in drug development, i.e. before drafting an IND proposal. It takes months to be covered in regulatory forms and delays how quickly a drug can hit the market. Yang began drafting the IND form last August, but ended in January 2025 and was approved. After submission, these applications will continue with Waylay Workers. An IND application is a “living” document. Additionally, in the 30-day evaluation period after submission, the FDA can request additional information or revisions if it finds a need and needs to delay the pipeline further.
Drug developers have strong incentives to promote IND form submissions while maintaining accuracy and thoroughness. With an AI (AI) program on increasing biology, developers want to try out AI-based solutions to accelerate this step. “If AI really helps to quickly summarise all the information and FDA regulatory requirements, it will go a long way in saving a lot of time in preparing the document,” he said it has not used AI assistance in its recent IND application.
Drug developers are beginning to use AI tools to employ some of the grunt works from the IND application to help draft text using large-scale language models (LLMs) like The ChatGPT. In fact, the Center for Drug Evaluation and Research, a division of the FDA, has found that between 2016 and 2023, more than 500 submissions in IND reports and other regulatory forms contain elements of AI. For example, for details on how drugs are manufactured, sponsors have found the best strategies to use AI to expand production.
Tools like CHATGPT help sponsors work through applications, but technology company Weave Bio has decided to create Autoind, the first AI tool designed to generate rough IND drafts within a day. According to Brandon Rice, Chief Product Officer of Weave Bio, IND applications can take up to six months to complete, but a quick draft using Autoind could save at least half the time.
AutoInd is a web application that can be used to upload all source documents as PDFs, whether sponsors are formalized reports or electronic research notes. “Essentially anything that can be made into a PDF can be taken as input,” Rice said. He suggested that users could quickly generate drafts of each IND form. “It's easy because you literally click one button per document.”
AI tools may prove attractive to startups who have never submitted IND proposals before. Nathan McMahon, clinical project manager at Trace Biosciences, is a company that manufactures fluorescent compounds that illuminate nerves during surgery and has no experience in IND application filing. Rather than helping consultants help draft IND documents, they gave Autoind a pivotal turn. “We generated 50 pages of content in an hour,” McMahon said. It only took two months to improve the application of fluorescent nerve probes. The team will submit an IND form soon.
AutoInd uses OpenAI LLM's constantly adaptive selection to extract key information from the text and tables (but not diagrams) in the source document and fill out the form. However, LLM has a reputation for creating errors, such as creating citations for non-existent scientific papers. These so-called “hastisation” occur when LLMS confuses information sourced from its own database. To work around this issue, Weave restricted LLM to information from uploaded documents. When McMahon generated the draft in Autoind, he only tweaked each document for an hour to tweak it to the loose ends before submitting the form to the consultant for review. “They were blind too. I didn't tell them how I made the content,” McMahon said. “The consultant came back and there were stylistic changes, but the content itself was accurate,” he added. Nevertheless, sponsors need to ensure that all data is correct, just like if a human drafted the form, Rice said. This makes the refinement stage the most time-consuming part of your application. “We plan to save a lot of time building the base, but we don't take away the actual monitoring and detailed reviews that team members need to do,” said Maria de Assis, vice president of clinical operations at Axcynsis Therapeutics.
Autoind is still under development, so there is room for improvement. For example, McMahon discovered that AI models are the most struggling in the manufacturing section. “There are so many documents involved, so it can be difficult to make all the models three and bring out important pieces,” explained McMahon. De Assis noted that it is beneficial when AI tools can help analyse preclinical data and propose clinical trial designs that follow FDA guidelines. “We did [the IND paperwork] It's been over a month and a half, but I think I'll have at least six months to get results and get it,” she said.
Some drug developers are worried about handing over data to AI. Yang pointed out that “many of Ind's information is confidential. How can I set up a firewall and make sure that the information is not leaked?” She pointed out that even AI developers who create tools for submitting Ind documents should not be able to access sensitive materials due to the risk that they could leak to competitors. “We are implementing best practices for all industry to ensure that our data is secure, encrypted and protected from access by others,” he added Weave Bio will have zero data retention agreements with Openai. “They agreed not to use the data they send to do their learning. In fact, they don't even keep it on the server,” Rice said.
Rice believes tools like Autoind are relatively error-free and safe, but the FDA wants to ensure sponsors can use AI responsibly. (FDA officials were unable to comment due to a White House-inspired suspension in external communications.) In January 2025, they published draft guidance for using AI at all stages of the drug lifecycle, including IND applications. They did not outline legally enforceable liability. However, the guidance provides recommendations on how to test the reliability of the AI models used by sponsors. For example, it suggests that sponsors need to quantify “model risks.” This is the possibility that the model produces inaccurate outputs, resulting in poor decisions and unfavourable outcomes for the trial participants.
Future AI programs may be equipped with additional features that simplify the submission of IND documents. For example, McMahon proposed that it could improve communication by summarizing data in a more concise and clearer way. There are many features that developers can add to the bot, but Rice has set his priorities. Sponsors are not hanging from their heads because they “want to streamline automatic content updates very much.”
AI aid hands allow drug developers to advance their pharmaceutical pipelines quickly. “Instead of me sitting there and typing everything and learning how to write it, I could spend more time moving forward on our clinical trial strategy,” McMahon said. AI is likely to have a growing presence in the pharmaceutical industry, and tools like this could allow developers to acquire new therapies at a record-breaking pace, from lab benches to bedsides.
