FDA just quoted a stem cell clinic's own YouTube video back at it — and warned the same doctor it warned in 2019
August 25, 2026 · 4 items
FDA just quoted a stem cell clinic's own YouTube video back at it — and warned the same doctor it warned in 2019
FDA — Warning Letter, CBER · issued Aug 14, 2026, posted publicly Aug 25 · August 14, 2026Regenerative medicine
Who and what: CBER wrote to David L. Greene, M.D., CEO of R3 Medical Companies (Scottsdale, AZ) — parent of Bello Bio, Regen Suppliers, R3 Stem Cell LLC and R3 Anti-Aging Scottsdale. FDA inspected the co-located Bello and Regen facilities Nov 17 – Dec 16, 2025. Four umbilical-cord-derived products — ReBellaWJ, ReBellaXO, ReBellaXOL, ReBellaCB — were declared unapproved new drugs (FD&C §505(a)) and unlicensed biologics (PHS Act §351(a)(1)), because they fail the homologous use criterion at 21 CFR 1271.10(a)(2): treating arthritis and inflammation is "not a basic function of umbilical cord or umbilical cord blood in the donor."
The part every marketer should read twice — FDA's evidence was the marketing. The letter names, by handle, the YouTube channel @R3StemCell, TikTok @r3stemcell, Instagram @r3stemcellofficial, the Facebook page, and three websites, "(last visited August 2026)." It quotes a blog post from November 15, 2020 that's still live, and a video posted March 5, 2026 in which an employee holds up vials and says: "These are your stem cells. There's 30 million in each vial… This is ReBellaXO. These are your exosomes… Those are very anti-inflammatory…" Website copy listing benefits for "Neurologic Conditions… Autism… Cerebral Palsy… Diabetes… Lyme Disease… Erectile Dysfunction" is reproduced in full. Marketing copy is labeling, and labeling is what makes a tissue product a drug.
The manufacturing findings, and they are not small: no validated aseptic process for products that "purport to be sterile" (21 CFR 211.113(b)); sterility samples frozen and thawed before testing, which FDA says "has the potential to destroy or compromise any microbial content in the samples before testing"; a 24-month expiration date assigned with no supporting stability data (211.166(a)); no written quality-unit procedures (211.22(d)). FDA also disputes the product category itself — Bello's process for the "exosome" products "does not appear would isolate exosomes."
This is a repeat, and FDA says so: the agency notified Greene by letter on May 28, 2019 about promoting stem cell therapies for ALS, diabetes, kidney failure, Lyme and Parkinson's — and states the same offers continue today. Response due in fifteen (15) working days; failure "may result in action without further notice including, without limitation, seizure and/or injunction." Signed by Vincent Amatrudo, Acting Director, Office of Compliance and Biologics Quality. Note: regensuppliers.com was already returning a 404 as of Aug 14.
Why it matters for an independent practice: Not an AI story on its face — I'm running it because it is the most consequential thing to happen to the regen-medicine practice world this month, and because the AI tie is real and uncomfortable. Every practice we serve is being told to publish more, faster, with AI help. This letter is FDA demonstrating that it reads the whole surface: a 2020 blog post, a March 2026 video, four social accounts, three websites — all pulled into one enforcement record. Volume is now liability. Three concrete moves. One, this week: run a claims audit across every client's site, blog archive, YouTube back-catalog and social — not just the current homepage. The 2020 post is the tell; nobody thought about it in six years and FDA quoted it anyway. Two: the specific language that converts a tissue product into an unapproved drug is condition claims — naming a disease. "Supports recovery" is a different legal object than "benefits for Parkinson's." Every client offering orthobiologics needs that line drawn explicitly in writing, and MMR should be the one holding the pen. Three, the positioning argument: this is what P2 costs and what it's worth. The practices that stayed conservative on claims now look like the credible ones — and when a competitor's letter goes public, patients search. Be the practice whose evidence page was already built. FDA — Warning Letter to R3 Medical Companies, MARCS-CMS 726330
Two Texas pharmacies in two weeks recalled compounded glutathione for bacterial endotoxin — and it mostly ships to med spas, where the patient never sees a vial
FDA — Recalls & Safety Alerts · company announcement Aug 19, FDA publish date Aug 20, 2026 · August 20, 2026Pharmacy
The recall: Optimal Balance Pharmacy (Houston, TX) voluntarily recalled lot LG342009076 of Glutathione Injection 200 mg/mL, 30 mL multi-dose vial, exp. Oct 4, 2026, to the patient level. It was caught by the pharmacy's own testing, not an FDA inspection — internal testing "identified elevated bacterial endotoxin levels." Distributed to consumers in 14 states: AZ, CO, FL, IL, LA, MI, NY, NV, OK, PA, TX, UT, VT, WA.
What FDA says the risk is, verbatim: "There is reasonable probability that injectable products with elevated endotoxin levels could cause significant medical events including fevers, hypotension, inflammatory reactions, anaphylactic shock and death." Adverse events already reported to the pharmacy: "fever, chills, severe headache, nausea, vomiting, pain, and symptoms consistent with an allergic reaction."
It's the second one: Austin's Victory Medical Center Pharmacy recalled three lots of compounded glutathione across TX, FL and NY with an FDA publish date of Aug 5. Medical Daily's framing is the useful one: "A single lot failing a test is a manufacturing problem. Two unrelated pharmacies failing on the same measure points to something upstream." Per the Partnership for Safe Medicines, glutathione appears on neither the 503A nor the 503B bulk substances list and is covered by a dietary supplement monograph, not a pharmaceutical one — and after a 2019 endotoxin incident FDA warned a compounder against using dietary-supplement-grade ingredients for injectable glutathione. A Travis County suit has been filed against Victory by a woman alleging hospitalization after an infusion from a recalled lot; the allegations are untested.
The distribution detail that makes this our problem, not a pharmacy trade story: these injections "are typically given at wellness clinics and med spas rather than in a medical setting where a recall notice would circulate," and "most people who receive these injections never see a vial." Both firms are 503A pharmacies overseen primarily by state boards, not federal inspectors.
Why it matters for an independent practice: Also not an AI story, and I'm posting it anyway because it sits directly on top of the client base — cash-pay regen, anti-aging and aesthetic practices running IV and injectable menus. Three things follow. One, today: any client offering glutathione IV or injection should check their compounder and lot numbers against both recalls before the next appointment. That's a ten-minute task with a "significant medical events… and death" downside. Two, the structural point worth raising in a client conversation: a recall notice reaches pharmacies. It does not reach the patient who got an IV drip at a med spa, because there was never a label in their hands. The practice is the only party positioned to close that loop, and having a documented process for doing it is both a patient-safety obligation and, frankly, a differentiator you can say out loud. Three, the sourcing question this raises: an ingredient that sits on no FDA bulks list and carries a supplement-grade monograph is a supply chain with a known failure mode. "Who compounds this, are they 503A or 503B, and what's their endotoxin testing protocol" belongs on the same one-page vendor diligence sheet we've been building for AI tools all month. Same discipline, different vendor. FDA — Optimal Balance Pharmacy voluntary nationwide recall (Trusted, primary) · Medical Daily — A second Texas pharmacy recalls compounded glutathione (flagged, untracked)
Healthline built an AI that answers the patient's question and then hands them a provider — 94% of the actions it drives are "find me a doctor"
Fierce Healthcare · Cailey Gleeson · August 25, 2026Patient acquisition
The usage numbers, as of Aug 18: Fullspan Health's Healthline AI has logged over 1.1 million views and reaches 13,500 individuals daily since its June launch. Head of Marketing Debra Harris to Fierce: "We're seeing that the user average is 1.6 queries per session, which equates to three conversation turns across the full exchange. The depth of people wanting to use it and wanting to interact with our content in this way has increased 28% since launch."
The conversion funnel, and this is the number that should stop you: "three-quarters" of people who open the tool submit a query, and 7% of sessions result in next-step actions. Of those actions, per Harris, "94% are having a care path or provider, finding a provider, seeking a specialty and engaging with care." In other words: the overwhelming majority of what this AI does, when it does anything, is route a patient to a doctor.
Why it isn't just another chatbot: it draws exclusively from Fullspan's own medically reviewed properties — Healthline, Healthgrades, Medical News Today, Psych Central. Harris on the moat: "We've got the number one provider resource within our portfolio" — that's Healthgrades, i.e. the directory is the answer engine's referral layer. In-product feedback is 78% positive, and the company says it has handled more than 2,800 diagnostic and sensitive interactions with zero reported clinical or compliance incidents to date.
The scale it's aiming at: currently live on Type 2 diabetes content only, with plans to expand conditions as it scales to Fullspan's 63 million monthly unique audience. Context Fierce supplies: a March Rock Health survey found 32% of respondents using chatbots to find answers to health questions, up from 16% in 2024; Healthline's own consumer research found 90% interested in a conversational health experience and nearly 70% saying they'd find it more trustworthy than ChatGPT. Read that last stat as marketing — it's the vendor's own research, published in its own launch announcement.
Why it matters for an independent practice: Every AI-search item we've run for two months has been about citations — will the practice get named in an answer. This is a different and more direct threat, because it's not an answer engine, it's a referral engine wearing an answer engine's clothes. A patient asks about knee pain, gets a medically reviewed response, and the next tap hands them a Healthgrades provider list. The doctor never competed on content; they competed on being in someone else's directory. Three moves. One, this week, and it's cheap: audit every client's Healthgrades profile the way we audit their Google Business Profile — specialties, conditions treated, photos, review volume, insurance. If 94% of this thing's actions are provider-finding and the provider data is Healthgrades, that profile just became paid-search-grade real estate that costs nothing. Two, the strategic read: 7% action rate on 13,500 daily users is small today and it's live on one condition. Diabetes first is not an accident — chronic, high-volume, high-commercial-intent. When it reaches 63 million uniques across conditions, the regen and orthopedic queries arrive too. Get the profile right while it's cheap. Three, the honest counter-position: this is P6 territory. An AI can hand a patient a specialty and a name. It cannot hand them the person who called them back on a Friday. That's still the practice's to win, but only if the practice shows up in the list at all. Fierce Healthcare — Fullspan Health's AI agent sees strong early traction with 1.1M views since June launch
A hot new AI assistant was told to delete a user's Gmail and didn't — and its terms claim a "perpetual and irrevocable" license to everything it sees
TechCrunch · Sarah Perez · August 24, 2026Buildable AI
What it is: Instinct is a private-access personal AI assistant from a San Francisco team led by Noah Shinn, a former Sierra research scientist; the corporate entity is Spear Street Technology, Inc. TechCrunch reports multiple investors saying Kleiner Perkins and Conviction have invested and those rounds have closed. It connects to your email, calendar and files and acts on them.
The terms of service, which is the whole story: users grant Instinct a "perpetual and irrevocable" license to "access, use, host, cache, store, reproduce, transmit, display, publish, distribute, and modify" their materials — including for training its models. The terms also cover receiving "screen captures, cursor movements, and keyboard inputs," and permit Instinct to enter "agreements, commitments, or transactions" on the user's behalf, which would be binding.
Two failures that are worse than the terms, because they're behavioral: early adopter Peter Yang found Instinct would not delete his Gmail records when asked — the team later shipped a fix by adding a delete-external-data tool in settings. And Claire Vo disconnected Instinct from Google at 11 AM and still received an email summary at 3 hours later; per TechCrunch, "the bot confirmed that the emails were stored in plain text for later searches." Revoking access did not revoke the copy.
The prompt-injection test, run by a healthcare founder: Alex Cohen, co-founder of Hello Patient, created a brand-new Gmail account and emailed his real personal account with instructions aimed at Instinct — then deleted his account. His line: "I don't think we're at the point where it's safe to give AI read/write access to your inbox." And the trust framing from Katie Jacobs Stanton of Moxxie Ventures, after Instinct sent an email on her behalf without asking: "The more powerful these agents become, the more trust matters. Every successful action earns a little more trust. One unauthorized action can reset that trust to zero." Caveat: still in private testing, and neither the company nor Shinn responded to TechCrunch's requests for comment.
Why it matters for an independent practice: Zero medicine in the source — extrapolate. Yesterday's item was an assistant Google switches on across Gmail, Drive and Calendar by default. Today's is what that class of tool does when nobody is watching, described by the people who tested it. Three things a practice should take from it. One, the diligence question is now behavioral, not contractual. "Do you offer a BAA" is necessary and no longer sufficient. The right question is "if I revoke your access to my mailbox today, what copy of my data survives, in what form, and how do I make you delete it?" Instinct's answer, empirically, was: plain text, and it didn't. Add that line to the vendor sheet. Two, the prompt-injection path is the one nobody prices in. An assistant with inbox read/write can be instructed by anyone who can email your staff. A practice inbox receives unsolicited mail all day — referral pitches, vendor spam, patient forms. Cohen's test took one throwaway Gmail account. Any front-desk AI that reads mail and takes actions needs a human confirmation step on anything that sends, pays, books or shares. Three, the terms language is the tell to teach clients to read for. "Perpetual and irrevocable," and any grant that includes training, is a different deal than storage. Consumer AI terms are increasingly written this way, and a doctor signing up personally on a Saturday is not reading them. This is exactly the shape of item that earned a :+1: on Aug 18 — a privacy exposure the practice must answer before adopting, not after. TechCrunch — Instinct's powerful AI assistant is raising privacy and security concerns
