AI Giant Anthropic Operates In-House Biology Lab to Tackle Hallucinations in Physical Experiments and Enhance Safety

Anthropic Launches In-House Biology Lab

Leading AI developer Anthropic has revealed that it is operating an in-house biology experimentation laboratory. This facility serves as a verification and testing hub to evaluate the biology-related capabilities, safety profiles, and potential risks of the company's AI models.

Evaluating AI Safety Through Physical Experiments

Rather than a product release, this initiative represents infrastructure building as part of the company's ongoing AI safety research. Traditionally, the reasoning capabilities of AI models have been evaluated primarily within simulation environments. By establishing an adjoining physical lab, Anthropic can now directly compare and verify actual experimental data against AI reasoning results. The goal is to more precisely identify and suppress inaccurate reasoning and "hallucinations" in the biological knowledge generated by AI.

Background and Future Outlook

Improving the accuracy of biological reasoning in AI models is critically important from the perspectives of academic research and public health. Anthropic plans to further strengthen the feedback loop derived from physical experiments, formulating safety measures to ensure AI mitigates biological risks. By combining digital evaluation with real-world physical validation, this initiative is expected to set a precedent for building safer and more reliable AI systems.