As of March 2026, the Food and Drug Administration (FDA) has authorized over 1,400 AI-enabled devices for market use. To help achieve this, the FDA has worked to adapt a framework originally built for static devices into one capable of governing adaptive, learning software.
Having advanced framework for marketed AI devices, the FDA is now turning its attention to AI inside live clinical trials. In April 2026, the FDA announced the Real Time Clinical Trial (RTCT) initiative. RTCT will allow real-time clinical trial data to be transmitted directly from hospital sites to the FDA and will use AI to analyze predefined signals on safety, efficacy, dosing, and adverse events to give FDA reviewers earlier insight into how a trial is progressing. The agency has already successfully initiated proof-of-concept trials, announced the pilot program to begin in August, and submitted a Request for Information (RFI) regarding the program’s design and implementation. The FDA has explicitly stated that the pilot is designed only for early-phase (Phase 1 and 2) trials.
This blog explores what guidance FDA has created for AI-enabled devices, why that framework doesn’t extend to clinical trials, and what needs to happen before RTCT launches.
The Existing Framework: AI-Enabled Medical Devices
For decades, the FDA’s authorization of medical devices has followed the same process: a device is reviewed, authorized, and remains unchanged. Any significant modification required submission from the device manufacturer. This system works for static devices, but not for AI, which depends on its ability to change.
The FDA has taken steps to improve how the agency handles authorization of AI-enabled medical devices, publishing two significant frameworks in 2025. The agency published the Total Product Life Cycle (TPLC) draft guidance which requires manufacturers to document how AI tools are taught, how bias is assessed throughout the building process, and how the tool will be monitored on the market. The FDA also finalized its Predetermined Change Control Plan (PCCP), which allows manufacturers to define anticipated AI modifications and get pre-authorization for those specific changes.
TPLC and PCCP create a solid regulatory environment for AI in medical devices, which can be anticipated and pre-authorized. The FDA has created a clear pathway for manufacturers and a structured system to monitor the devices. The framework is not complete, TPLC remains a draft, but there is real progress. However, this work was designed on the assumption that a manufacturer can control and anticipate the changes its AI will make. RTCT breaks that assumption entirely.
The New Frontier: AI in Live Clinical Trials
Clinical trials occupy a fundamentally different regulatory space than devices. They are governed by the Center for Drug Evaluation and Research (CDER), have different statutory authorities, and have different stakeholders.
RTCT would change how FDA reviewers access and analyze trial information. Rather than waiting for submitted data packages, reviewers would receive predefined signals, analyzed by AI, as the trial is actively running. The FDA issued draft guidance in January 2025 on the use of AI to support regulatory decision-making for drugs and biologics. This is the closest analog to what RTCT would do, but it remains a draft more than a year later with no finalization timeline announced.
The frameworks built for devices do not apply here, as they were created for manufacturers who control their AI systems. RTCT involves AI operating on live trial data to analyze signals that no existing regulatory framework governs.
Concerns with RTCT
As the FDA develops the regulatory framework for AI in clinical trials, stakeholders have four critical questions regarding how RTCT will actually function.
Accountability: Who Is Responsible When AI Gets It Wrong?
Under RTCT, AI-analyzed signals will directly inform consequential reviewer decisions, including whether to stop a trial arm, change a dose, or accelerate enrollment. Yet no existing policy assigns responsibility when an AI-generated signal is flawed and a resulting decision causes patient harm. The answer to this question matters for determining who has oversight and the extent to which there is public trust in this initiative.
Bias: How will RTCT address concerns about bias?
RTCT will process live data continuously from a small number of academic medical centers, where data may not reflect national diversity. Bias can emerge dynamically in the data as it accumulates in ways a pre-trial assessment cannot detect. There is no existing policy that requires ongoing bias monitoring during a live trial, and there is no infrastructure to mandate or receive such reporting.
Auditability: Will there be an opportunity for stakeholders to challenge RTCT decisions?
There is currently no requirement that AI signal outputs be logged in a form that can be reviewed, explained, or challenged after the fact. Without a mandatory audit trail, stakeholders have no mechanism to challenge or reconstruct decisions that shaped a drug’s approval pathway. At the May 15 stakeholder session, FDA officials acknowledged that data cleaning, governance, and quality-control requirements for the signal stream have not yet been finalized.
Success Metrics: How will the FDA know if RTCT is working?
A fourth concern emerged directly from the May 15 stakeholder session as the FDA has not yet defined how it will measure whether RTCT actually improves decision-making. Launching a pilot without predefined performance criteria makes it difficult to evaluate whether the program should be expanded or stopped.
RTCT Comment Period
The April 28th RTCT announcement included a request for comments on how the initiative should be designed and evaluated. Comments received to date reflect the concerns above with stakeholders urging the FDA to:
- Require capture of underrepresented populations in AI training data
- Establish version-traceable lineage so that model changes can be audited
- Incorporate patient advocate oversight into RTCT governance structures
- Review site-level operation data to monitor for performance disparities across sites
- At the May 15 session, stakeholders also requested an extension to the comment period, but the FDA declined to commit to an extension.
What’s Next?
As the comments demonstrate, actually creating regulations for the RTCT is a difficult process. There is one additional variable that cuts across all of the above: FDA capacity. The agency has lost nearly 15% of its workforce since 2023 and is operating without a confirmed commissioner. Adding to the uncertainty, Jeremy Walsh, FDA’s first-ever chief AI officer and the official who led the May 15 session, has since resigned from the agency. The RTCT pilot is an ambitious project for an agency that is being asked to do more with considerably less, including the person who architected the program.
Congress has also taken notice but has not yet acted. The House Energy and Commerce Health Subcommittee held hearings in 2025 examining AI in health care and the Senate HELP Committee flagged FDA modernization as a priority. However, the hearings have not produced legislation, and the administration’s own National Policy Framework for AI, released in March 2026, is only a blueprint for Congress.
The most consequential near-term legislative vehicle would be one of the FDA user fee reauthorization bills. Congressional authorization for both the Medical Device User Fee Amendments (MDUFA) and the Prescription Drug User Fee Act (PDUFA) expires on September 30, 2027. These programs fund the FDA’s pre-market review operations and are widely considered “must-pass” legislation. Congress has historically used these reauthorization bills to enact broader FDA policy reforms, meaning these could codify AI regulatory authorities for medical devices and clinical trials. Negotiations between the FDA and industry are already underway, meaning stakeholder engagement in that process starts now.
Conclusion
The FDA’s progress on AI-enabled medical devices is real and has given the device industry a clearer regulatory path and agency oversight model. The clinical trial space is a different story. RTCT represents one of the most ambitious uses of AI in clinical research and if it works correctly, has the potential to vastly compress drug development timelines. But the risks are real. AI-analyzed signals will directly inform consequential regulatory decisions without finalized standards for accountability, bias monitoring, auditability, or success measurement. The official who built this program has resigned, and no successor has been named. The existing frameworks, TPLC and PCCP, were not designed for this case and the draft guidance closest to it remains unfinalized. Regulatory rules for AI in clinical trials are being written now. Stakeholders who engage in this process will help write those rules and those who wait will inherit them.
The RTCT public comment period closes May 29, 2026. Comments can be submitted here.