Researchers at Johns Hopkins University trained a system called SRT-H on videos of surgeons performing gallbladder removals, then set it loose on ex vivo pig tissue to carry out a key phase of that procedure across eight separate trials with no hands on the controls.
SRT-H completed the clipping and cutting phase on its own in all eight trials, adapting to each piece of tissue's slightly different anatomy, correcting itself when something did not go as expected, and it can also take spoken corrections from a human supervisor, though none were needed in these trials.
Ex vivo pig tissue sitting on a table is a controlled proxy for the mess of a living patient, and SRT-H has not yet operated on a living animal or a person. Built on the same kind of architecture that powers large language models, the system's developers describe it as a step toward autonomy, not a finished autonomous surgeon, with validation in live surgery still ahead.

Read the original study
Kim et al., Science Robotics, 2025 · doi.org/10.1126/scirobotics.adt5254
Eight operations, zero hands on the controls.
Source
- Kim, J. W. B., Chen, J.-T., Hansen, P., Shi, L. X., Goldenberg, A., Schmidgall, S., Scheikl, P. M., Deguet, A., White, B. M., Tsai, D. R., Cha, R. J., Jopling, J., Finn, C., & Krieger, A. (2025). SRT-H: A hierarchical framework for autonomous surgery via language-conditioned imitation learning. Science Robotics, 10(104), eadt5254. https://doi.org/10.1126/scirobotics.adt5254