FDA issued a discussion paper on November 6, 2026, titled "Evaluating Benefit-Risk in Robotic Medical Devices with Autonomous or Remote Teleoperation Capabilities," asking for public feedback ahead of a hybrid workshop set for December 2 and 3. The agency is accepting comments through January 5, 2027, and says the paper is not proposed guidance, only a basis for the workshop discussion.
The paper treats cybersecurity and autonomy as compounding risks rather than separate categories. FDA wrote that at higher levels of autonomy, "device-generated errors may be initiated and executed before a supervising user can meaningfully evaluate or interrupt them," and it raised the specific example of a remote vascular procedure losing real-time imaging mid-procedure, which could let a robotic instrument enter the wrong vessel, perforate a vessel, or dislodge a clot before anyone catches it.
As of July 2026, FDA said no remotely teleoperated or fully autonomous robotic device had received US marketing authorization, though conventional robotic surgical systems and devices with limited autonomy already have. Developers are pursuing devices that draw blood without a phlebotomist, that find and stop internal bleeding without a clinician's input, and remote ultrasound and telesurgery systems. Urologist Vipul Patel performed a robotic prostatectomy in October on a patient roughly 35 miles away under an investigational device exemption, a year after performing one across the Atlantic on a patient in Angola.
Outside experts who reviewed the paper flagged gaps. Kevin Fu of Northeastern University's Archimedes Center asked what happens if "AI-powered attacks manipulate surgical trajectories to make deliberate harm look like an accident," and called for graceful degradation rather than global outages when something goes wrong. Phil Englert of Health-ISAC said the communication link between operator and device is now part of the attack surface and argued for cyber-informed clinical risk assessments alongside physical emergency-stop capability. Axel Wirth of MedCrypt said the paper treats AI and cyber risk separately and does not address adversarial machine learning, model integrity, or update-channel security, nor does it divide accountability among remote clinicians, hospitals, telecom providers, and cloud services.
The paper is non-binding, but it is the clearest signal yet that FDA intends to fold telecom grade failure modes, latency, jitter, packet loss, corrupted video, into the same benefit-risk framework it already uses for AI reliability and cybersecurity. Teleoperated and autonomous device submissions should expect FDA to ask about all three together rather than file them under separate review tracks.