Generation Patient comments on the MDUFA VI draft commitment letter

Written By Sneha Dave
Generation Patient comments on the draft MDUFA VI commitment letter

Every five years, the Food and Drug Administration (FDA) negotiates an agreement with the medical device industry over the fees companies pay to have their products reviewed and the performance goals the agency promises in return. Those terms are set out in a commitment letter under the Medical Device User Fee Amendments (MDUFA), and Congress relies on that agreement when it reauthorizes the program. In July 2026, the FDA published their draft commitment letter for the next MDUFA cycle and invited stakeholders to file comments with views about the agreement.

Download the Generation Patient comments on the MDUFA VI draft commitment letter

Last week, Generation Patient delivered an oral statement at the FDA about this draft commitment letter and filed formal comments. Read the Generation Patient comments on the MDUFA VI draft commitment letter below.

August 7, 2026

Food and Drug Administration
5630 Fishers Lane, Room 1061
Rockville, MD 20852

Re: FDA-2026-13778: Medical Device User Fee Amendments

Generation Patient is an organization created by and for young adult patients living with chronic and rare medical conditions such as lupus, inflammatory bowel disease, Lyme disease, tuberous sclerosis, and rheumatoid arthritis. We seek to ensure a better future for our generation of patients by providing direct support through peer support groups while driving systems-level change through policy work, leadership programming, and advocacy initiatives. Through our direct support work, we have led over 650 peer support meetings and continue to build the evidence base to build holistic support for young adults with chronic and rare conditions.

Generation Patient asks FDA to make the following commitments in the final MDUFA VI commitment letter:

  1. Commit to strengthen oversight of AI companion tools and chatbots based on their design features and potential to harm users, rather than on manufacturers' marketing claims.
  2. Commit CDRH to publish draft guidance on generative AI in digital mental health devices within the first year of MDUFA VI, covering the evidence needed to demonstrate safety and effectiveness and crisis response performance.
  3. Explain how FDA will handle updates to the foundational models underlying AI devices like companion tools and chatbots, including monitoring and recalls.
  4. Add postmarket surveillance commitments with performance goals as specific as the review goals, and dedicate increased user fee funding toward postmarket studies for at least class II and III devices.
  5. Disaggregate adverse event data by patient age, beginning with a bracket spanning ages 18 through 29, so that young adults remain visible in device safety data.
  6. Finalize the September 2023 draft guidance on predicate device selection within the first year of MDUFA VI, and provide that a device without a valid predicate proceeds through a pathway that evaluates it on its own evidence.
  7. Establish a presumption against clearance for any 510(k) submission citing as predicate a device subject to a Class I recall or an unresolved Section 522 order, with automatic flagging of such submissions.
  8. Extend to patient, consumer, and public health organizations the annual right industry holds under Section IV.C to propose review consistency topics, and prioritize predicate quality and AI-enabled devices among those topics.
  9. Report 510(k) predicate characteristics in the Section V performance reports, and dedicate user fee funding to modernize the 510(k) database so patients can trace the full predicate chain of any cleared device.
  10. Clarify that generative AI products offered for mental or behavioral health will not be cleared through the 510(k) pathway, will instead be evaluated on their own clinical evidence under the premarket review section 513(f)(2) for novel devices, and will not serve as predicates for future 510(k) submissions.
  11. Publish complete minutes of FDA and industry negotiation meetings, respond in writing to public comments before transmitting the agreement to Congress, and extend recurring consultation rights to patient, consumer, and public health organizations.

1Strengthen oversight of artificial intelligence companion tools and chatbots

Nine months ago, the Food and Drug Administration (FDA) convened its Digital Health Advisory Committee on generative artificial intelligence (AI) in mental health.1 Sneha Dave, our Executive Director, testified at the meeting. Experts explained that companion tools and chatbots can foster isolation and dependency, and can fail to identify crises that need escalation, including suicide.2 Clinicians and developers testified that sycophancy in these systems is a design feature rather than a defect, with an industry representative describing it as “almost required by design” in consumer products.3 Researchers testified that large language models can reinforce delusions, and that many chatbots falsely claim to be licensed therapists.4

In the Digital Health section, the draft commitment letter commits FDA to “develop software and digital health technical expertise to provide assistance for premarket submissions that include software, interoperable devices, or otherwise incorporate digital health technologies, such as generative and agentic artificial intelligence (AI) […].” However, the draft commitment letter fails to make any references to mental health, to crisis, to suicide, to chatbots, or to companion tools. These are essential topics to the millions of young adults with chronic conditions across America.

Generation Patient calls for concrete commitments to increase FDA oversight of AI companion tools and chatbots. Products that simulate a relationship, engage users in distress, and present themselves as emotional support function as mental health interventions even if manufacturers describe them as wellness products. FDA oversight of AI companion tools and chatbots should therefore be based on their design features and potential to harm users, including young adult patients, rather than on manufacturers' marketing claims.

Generative AI introduces new considerations for study design, including appropriate control arms, blinding, outcome measures, and the timeframe needed to demonstrate meaningful clinical benefit.5 However, FDA guidance on clinical evidence for digital mental health devices remains under development.6 The final letter should commit the FDA Center for Devices and Radiological Health (CDRH) to publish draft guidance on generative AI in digital mental health devices within the first year of MDUFA VI, covering the evidence needed to demonstrate safety and effectiveness and crisis response performance. Some of the questions that such guidance must answer include what constitutes a reasonable control arm for a chatbot trial, what rate of false-negative for safety events is acceptable, how therapeutic consistency is measured for AI systems that generate different output for every user, and how crisis-escalation behavior is tested before FDA clearance.

Predetermined change control plans (PCCPs) can govern changes made by manufacturers to their own medical devices.7 However, most conversational AI tools are built on top of a foundational model. When developers update foundational models, downstream medical devices – and their behavior – change. If the model is supplied by a third party, the device manufacturer neither makes nor controls that update. Models can also “drift” and adopt previously unseen behavior. Researchers testing the same system two months apart have documented a complete reversal in behavior.8 The draft letter commits FDA to build review capacity for PCCPs but fails to address updates in foundational models. The final letter must explain how FDA will handle updates in the models used for AI devices like companion tools and chatbots, including monitoring and recalls.

2Allocate user fees to support expanded post-market surveillance activities

Young adults who are using medical devices deserve stronger post-market surveillance of these products. A patient in our community who receives an implanted device or begins relying on a cleared product at age twenty-two will live with the consequences of weak surveillance longer than most adult users.

Across thirty-six pages, the word “postmarket” appears only once in Section IV.H.4.b, related to digital health, where FDA commits to “promote approaches that balance premarket and postmarket evidence […].” The terms “surveillance,” “adverse event,” “medical device report,” “recall,” and “safety signal” appear zero times in the draft. The term “post-approval” appears once, in a paragraph governing major deficiency letters. Section V specifies every metric FDA will report over five years, listing review times, review cycles, acceptance rates, withdrawal rates, hiring counts, and fee collections. Section V makes no mention of adverse event volume, recall count, safety signals, or patient outcomes. Commitments about medical device reports, Section 522 orders, signal detection, or recall analysis are also absent from the draft letter.

Faster clearance without corresponding surveillance capacity shifts the burden of harm onto patients. Section I.B commits FDA to reduce the average total time to decision for 510(k) submissions from 128 calendar days in FY 2028 to 112 days in FY 2032. Meanwhile, Section IV.H acknowledges “rapid growth and complexity of digital health topics in regulatory submissions,” naming generative AI tools. The agreement therefore accelerates review of an increasingly novel device population while committing nothing to “postmarket” surveillance after FDA clearance.

Generation Patient asks FDA to add postmarket surveillance commitments to the final letter and to dedicate increased user fee funding toward requiring postmarket studies for at least class II and III devices. Every goal in the draft measures how quickly FDA decides applications, and none measures how quickly FDA detects harm. FDA should commit to postmarket performance goals that match the level of specificity described in the review goals. Moreover, adverse event data should be disaggregated by patient age, so that young adults remain visible in device safety data rather than absorbed into a single bracket spanning ages 18 through 64. Generation Patient proposes that the Adverse Event Monitoring System (AEMS) reports use the following brackets: 18 through 29 years; 30 through 39 years; 40 through 49 years; 50 through 59 years; 60 through 64 years; 65 through 74 years; 75 years through 85 years; and over 85 years.

3Reform the 510(k) pathway and strengthen predicate device standards

Clearances made through the 510(k) pathway rest on chains of earlier decisions. Under this pathway, applicants can secure FDA clearance by showing “substantial equivalence” to a “predicate” device. The integrity of the 510(k) pathway therefore depends on which devices are used as “predicates,” the evidence FDA used to clear those earlier products, and the rigor of post-market surveillance efforts led by the FDA to ensure they remain safe. However, evidence around the 510(k) pathway suggests that this pathway is deeply flawed and experts have called for reform.9 Researchers examining 156 devices subject to Class I recalls between 2017 and 2021 found that, among the 127 with identifiable predicates, 44.1 percent cited predicates that had undergone Class I recalls, and that 119 of the 156 underwent no premarket clinical testing.10

FDA has already acknowledged this problem. In September 2023, the agency published draft guidance on best practices for selecting a predicate device, recommending that applicants avoid predicates with unmitigated safety issues or design-related recalls.11 Nearly three years later, that guidance remains a draft. The final MDUFA VI commitment letter should commit FDA to finalize the predicate selection guidance within the first year of the agreement. The final guidance should also close a loophole in the draft. Where no valid predicate consistent with the best practices exists, the draft permits applicants to describe how known concerns were mitigated. A device without a valid predicate is not substantially equivalent to anything and should proceed through a pathway that evaluates it on its own evidence.

FDA should commit in the final letter to establish, through rulemaking or binding policy, a presumption against clearance for any 510(k) submission that cites as predicate a device subject to a Class I recall or an unresolved Section 522 order. An applicant relying on such a predicate should be required to submit clinical data. Review systems should automatically flag these submissions, and user fee funding should support building that capability into FDA's premarket review infrastructure.

Generative AI strains the predicate flaws further. FDA evaluates applications under the 510(k)-pathway based on predicate devices as they existed at the time of clearance.12 Yet AI devices can be modified under a predetermined change control plan or through foundational model updates. The behavior of generative AI devices can change substantially following those updates.13 A later 510(k) applicant citing those tools as a “predicate” will compare their product with a set of specifications that the marketed device may no longer match. Furthermore, for conversational generative AI products that produce different output for every user, comparing a 510(k) submission to a marketed tool, establishes nothing about how the new device will behave. None of these challenges are clearly addressed by FDA in the draft MDUFA VI commitment letter.

Section IV.C of the draft letter commits FDA to improve review consistency in at least one specific, high-impact topic area each fiscal year, with a minimum of eight topics across five years. Each fiscal year, “industry will provide FDA with a prioritized list of specific, high-impact topic areas.” The draft letter does not name any priority topic and fails to grant any stakeholder besides industry similar rights to propose them. Generation Patient asks that FDA extend the same annual submission right to patient, consumer, and public health organizations, which bear the consequences of inconsistent review. FDA should also commit to select the quality of products used as “predicate” for 510(k) clearance and the consistency of review for devices enabled by AI, including companion tools and chatbots, among the priority topics addressed during MDUFA VI.

FDA agreed to make “performance reports” in Section V of the draft letter, including several metrics related to the 510(k) pathway. Quarterly reporting includes average FDA days, industry days, and total days to decision, average number of review cycles, and the rate of submissions not accepted for review. Yet none of the measures listed in the draft letter acknowledge the deep flaws of the 510(k) pathway and predicate selection practices. Left unmodified, the "performance report" that FDA committed to under Section V will measure only how quickly submissions move and will tell the public and young adult patients across America nothing about the quality of the “predicate” decisions those submissions rest on.

Generation Patient asks FDA to report 510(k) predicate characteristics for each cohort, including the average number of predicates cited, the share of clearances citing a predicate subject to an unresolved recall, and the share citing a predicate that was itself cleared without clinical data. FDA should also report performance disaggregated by whether a 510(k) submission involves generative AI or cites as “predicate” a device enabled by generative AI.

Patients currently have no practical way to trace the predicate lineage of a device they use. FDA should dedicate MDUFA VI user fee funding to modernize the 510(k) database so that every clearance displays its full predicate chain, including the recall status and clearance basis of each device in that chain. A young adult choosing between two continuous glucose monitors should be able to see whether either rests on a recalled predicate.

Generation Patient further asks that FDA state, in the final commitment letter and agency guidance, that it will not clear generative AI products offered for mental or behavioral health through the 510(k) pathway. For products whose output differs for every user, a comparison to a marketed predicate cannot demonstrate substantial equivalence, and clearance on that basis would begin a predicate chain resting on minimal safety and effectiveness evidence from the first submission. These products should instead be evaluated on their own clinical evidence, including crisis response performance, under the premarket review that section 513(f)(2) of the Federal Food, Drug, and Cosmetic Act provides for novel devices. FDA should further commit that no generative AI product offered for mental or behavioral health will serve as a predicate for any future 510(k) submission.

4Ensure meaningful patient participation in the MDUFA process

Generation Patient has participated in every one of the MDUFA VI public stages required in Section 738A of the Federal Food, Drug, and Cosmetic Act, 21 U.S.C. § 379j-1(b). However, the core bargaining over fee levels and performance commitments has occurred in bilateral sessions between FDA and industry, without the presence of independent organizations like ours.14 This is reflected in the draft agreement. Across thirty-six pages, the draft letter mentions “patient organization,” “patient group,” and “consumer” zero times. The draft commitment letter grants industry at least nine recurring consultation rights, including the right to provide annual input into audit objectives under Section III.A and to elevate concerns to FDA leadership within seven days under Section III.B. Patients appear in the draft as one example of stakeholders FDA may engage through unspecified mechanisms under Section IV.H, and as beneficiaries of the Total Product Life Cycle Advisory Program (TAP).

Generation Patient asks FDA to publish complete minutes of negotiation meetings with industry rather than brief summaries.15 FDA should also respond in writing to public comments on the draft commitment letter before transmitting the agreement to Congress, identifying which recommendations were adopted and which were not. FDA should further extend to patient, consumer, and public health organizations the recurring consultation rights the draft letter grants industry, including participation in performance meetings and the annual engagement on digital health topics under Section IV.H.

FDA exists to protect public health. The draft commitment letter does not yet meet that standard. Generation Patient will continue working to ensure that the final agreement does, and that medical devices meet rigorous evidentiary requirements both before and after they reach patients. We look forward to continued collaboration to ensure that medical devices are safe and effective for the current and next generation of patients.

Luis Gil Abinader
Policy Director
Generation Patient
luis@generationpatient.org

Sneha Dave
Executive Director
Generation Patient
sneha@generationpatient.org

Notes

  1. U.S. Food & Drug Admin., Ctr. for Devices & Radiological Health, Digital Health Advisory Committee (DHAC) Meeting, Brief Summary on the Topic “Generative Artificial Intelligence-Enabled Digital Mental Health Medical Devices” (Nov. 6, 2025), https://www.fda.gov/media/190450/download ↩
  2. Id. at 5–6, 14 (recounting testimony describing “unsafe responses to suicidal users,” “AI psychosis,” and “particularly unsafe responses to delusion and suicide-related prompts,” and committee concerns regarding “potential overuse or dependence.”) ↩
  3. Transcript of Digital Health Advisory Committee Meeting, U.S. Food & Drug Admin., at 04:43:54 (Nov. 6, 2025) (statement of Andrew Trister) (“there's a level of sycophancy that you could see in the consumer world it's almost required by design not really a bug”) ↩
  4. Id. at 5–6 (noting that "many chatbots falsely claim to be licensed therapists" and presenting "evidence that LLMs can reinforce delusions"). ↩
  5. FDA, Brief Summary, Digital Health Advisory Committee Meeting, November 6, 2025, at 2, available at https://www.fda.gov/media/190450/download (Pamela Scott stating that “GenAI therapeutics introduce new considerations for study design, including appropriate control arms, blinding, outcome measures, and the timeframe needed to demonstrate meaningful clinical benefit.”) ↩
  6. U.S. Food & Drug Admin., Ctr. for Devices & Radiological Health, CDRH Proposed Guidances for Fiscal Year 2026 (FY 2026), https://www.fda.gov/media/188993/download (listing "Clinical Evidence Considerations for Digital Mental Health Treatment Devices, including Computerized Behavioral Therapy Devices" on the Under Construction List) ↩
  7. Federal Food, Drug, and Cosmetic Act § 515C, 21 U.S.C. § 360e-4 (authorizing predetermined change control plans for changes a manufacturer plans to make to its own approved or cleared device); U.S. Food & Drug Admin., Marketing Submission Recommendations for a Predetermined Change Control Plan for Artificial Intelligence-Enabled Device Software Functions (Aug. 2025), https://www.fda.gov/media/166704/download. ↩
  8. Peter Kirgis, Ben Hawriluk, Sherrie Feng, Aslan Bilimer, Sam Paech & Zeynep Tufekci, LLM Spirals of Delusion: A Benchmarking Audit Study of AI Chatbot Interfaces, arXiv:2604.06188 (Feb. 20, 2026), https://arxiv.org/abs/2604.06188 (finding that "the same API endpoint tested just two months apart yields a complete reversal in behavior"); see also Lingjiao Chen, Matei Zaharia & James Zou, How Is ChatGPT's Behavior Changing Over Time?, 6 Harv. Data Sci. Rev., no. 2, 2024, https://doi.org/10.1162/99608f92.5317da47. ↩
  9. Inst. of Med., Medical Devices and the Public's Health: The FDA 510(k) Clearance Process at 35 Years 193, 196–97 (2011) (calling FDA to develop “a new medical device regulatory framework for Class II devices so that the current 510(k) process... can be replaced with an integrated premarket and postmarket regulatory framework”) ↩
  10. Kushal T. Kadakia, Sanket S. Dhruva, César Caraballo, Joseph S. Ross & Harlan M. Krumholz, Use of Recalled Devices in New Device Authorizations Under the US Food and Drug Administration's 510(k) Pathway and Risk of Subsequent Recalls, 329 JAMA 136 (2023), https://doi.org/10.1001/jama.2022.23279. ↩
  11. U.S. Food & Drug Admin., Best Practices for Selecting a Predicate Device to Support a Premarket Notification [510(k)] Submission, Draft Guidance for Industry and FDA Staff (Sept. 2023), Docket No. FDA-2023-D-3134. ↩
  12. Federal Food, Drug, and Cosmetic Act § 513(i), 21 U.S.C. § 360c(i). ↩
  13. Lingjiao Chen, Matei Zaharia & James Zou, How Is ChatGPT's Behavior Changing Over Time?, 6 Harv. Data Sci. Rev., no. 2, 2024, https://doi.org/10.1162/99608f92.5317da47 ↩
  14. Sneha Dave & Rita F. Redberg, A Primer of Medical Device Regulation: How to Go from Reactive to Proactive to Better Protect Patient Safety, (forthcoming 2026) (manuscript at 15–16). ↩
  15. See Therese J. Ziaks, Jason L. Schwartz, Joseph S. Ross & Reshma Ramachandran, Reforming the Prescription Drug User Fee Program, 393 New Eng. J. Med. 734, 734–35 (2025); Vinay K. Rathi, Joseph S. Ross, Rita F. Redberg & Sanket S. Dhruva, Medical Device User Fee Reauthorization — Back to Basics or Looking Ahead?, 387 New Eng. J. Med. 196, 196 (2022). ↩
Sneha Dave
Next

The Power of Peer Support: Getting to Know Our Peer Facilitators