OpenTox Virtual Conference 2026

People often ask: can you give a number to quantify the safety of this drug target? Toxicologists often respond: 50 numbers are OK, but one is hard, because safety is a multi-dimensional balancing act.

Both may agree: bringing safety evidence forward in target assessment is important. There are various ways to do this, including the increasingly popular use of AI agents. Agents offer the opportunity to ask almost any question, including the safety score question above. The issue isn't that AI can't answer that question. The issue is that it will!

This shifts the topic from simply finding an answer to asking the right question, using the right evidence, and knowing whether the answer is one a toxicologist would trust. How hands-on should a toxicologist - or indeed any human - be? 

Researchers are taking different approaches, from integrating safety data into carefully designed systems, to AI Scientists reasoning autonomously. What's constant is the demand for safety evidence that can be trusted by whichever consumer sits on top of it.

In this talk we share what Sable has built in practice with toxicologists and discovery biologists: a multi-domain data layer, summaries based on aggregated evidence, and tooling that lets scientists define scoring rubrics suitable for their indications. We made selected safety capabilities available to AI agents, based on preclinical safety teams’ recommendations. We’ll share what we’ve learned from these experiences: where AI can help, where guardrails matter, and where the data layer is most critical.

Agents may make answers easier to find, but they don't remove the need for human judgement. We still need to decide whether we asked the right question, whether the evidence supports the answer, and whether it reflects sound toxicological judgement.