ARPA-H put $98.5M behind the idea that rare-disease AI is failing on data plumbing, not on algorithms
September 3, 2026 · 5 items
ARPA-H put $98.5M behind the idea that rare-disease AI is failing on data plumbing, not on algorithms
Healthcare IT News · Andrea Fox, Senior EditorPractice operations
Four contracts, valued up to $98.5 million total over nearly five years, announced this week under a new ARPA-H program called RAPID — Rare Disease AI/ML for Precision Integrated Diagnostics.
Read what the money is actually for: "unifying and standardizing dispersed data and developing scalable approaches to safe data sharing, consent management and interoperability" to produce an AI-ready, real-world data resource linking longitudinal clinical records with genomic, patient-reported and other sources. Not a better model — the substrate underneath one.
The four awardees split the problem cleanly. University of North Carolina builds the largest real-world rare-disease dataset as a privacy-preserving foundation. Sage Bionetworks builds a platform for transparent performance assessment, common evaluation standards, and benchmarking of new AI approaches. FDNA builds tools to collect longitudinal health data at national scale — photos, videos, voice and patient-reported concerns. Probably Genetic generates high-fidelity synthetic datasets and patient-facing tools pulling from photos, videos, wearables, functional assessments, clinical records and genomics.
Scott Gorman, RAPID program manager: "Rare diseases represent one of the greatest unmet needs in medicine, and one of the most important frontiers for AI… By building AI-ready datasets and infrastructure, along with technologies designed to learn from sparse and complex rare disease data, RAPID aims to improve outcomes for patients while also advancing precision medicine more broadly." Stated goals: cut time to diagnosis, and enable more precise cohort identification for trials and drug development.
Why it matters for an independent practice: The federal government just made a $98.5M bet on where clinical AI actually breaks, and the answer wasn't intelligence — it was sparse, siloed, non-standardized records. That's the same wall a practice hits the moment it tries to do anything with AI beyond drafting: your ambitions are capped by your data hygiene, and no vendor fixes that for you. Second, quieter read: look at the collection surface being normalized here — photos, video, voice, wearables, patient-reported outcomes. That is precisely what a regen practice already generates at every visit and mostly discards. The agencies are building infrastructure to make that kind of messy longitudinal data usable. The practices that start keeping it in a structured way now will have something in three years; the ones that don't will still be reconstructing outcomes from chart notes.
A startup is selling AI models with the safety training surgically removed — and a reporter got malware and a pathogen protocol out of it in minutes
TechCrunch · Rebecca Bellan · September 3, 2026Buildable AI
Abliteration.ai hosts modified open-weight models with their refusal behavior stripped out — including Z.ai's recently released GLM-5.3 — queryable from a web browser or through an API. "Abliteration" is the name of the technique that removes a model's tendency to refuse harmful requests; the startup turned it into a hosted service, so users no longer need to download pre-abliterated weights and secure their own compute.
Their stated purpose, from a recent social post: to enable "offensive cyber, red-teaming, and agent testing work other models refuse to do." TechCrunch grants the logic is familiar in security work — you can't defend against a behavior you can't reproduce.
What TechCrunch actually did: created an account and started querying an abliterated GLM-5.3 for free through a web browser. They asked it to write a Python program that steals saved Chrome passwords, and for a detailed protocol for culturing a dangerous human pathogen at home. It readily complied.
The guardrails that remain are thin and self-imposed. Co-founder Devon (last name withheld at his request — he's still employed elsewhere) says the platform has some minor limits: in testing, TechCrunch couldn't get suicide instructions. There is no KYC beyond logging the credit card used to buy the service. The company has deals with major cloud providers funded purely by customer revenue, has taken no venture capital yet, and is in talks to. Andrew Yoon, head of research at AI-safety nonprofit CivAI, has argued governments should require providers to run classifiers detecting harmful cyber and bioweapons activity, and require companies renting advanced GPU access to verify customer identities and "deny access where there is reason to suspect dangerous misuse."
Why it matters for an independent practice: This is the "who can reach what, and under whose rules" shape that has earned :+1: here twice, and it lands on a comfortable assumption. When a practice buys an AI receptionist, an intake bot, or a content tool, everyone quietly assumes the safety behavior is a property of the model. It isn't — it's a product decision that a third party can remove and resell for the price of a credit card. So the vendor question sharpens: which model, hosted where, and has anyone modified it? A reseller who can't answer that in one sentence hasn't thought about it. The second-order effect is closer to home than the bioweapon headline: uncensored models make convincing patient-facing phishing, fake portal messages and physician impersonation cheaper to produce at volume — the same threat that hit Epic MyChart last week, with the production cost taken out.
"Fix this AI output" is now its own job market — cleanup listings up 87%
Search Engine Journal · Matt G. SouthernPatient acquisition
Freelancer.com internal platform data, reported by The Guardian today: job listings for correcting AI-generated work rose 87% from August 2025 to June 2026, totaling 10,760 posts globally. The listings are tagged with phrases like "correct AI," "AI hallucination," and "AI error."
Two corroborating signals from different platforms: Upwork reported a 70% year-over-year rise in AI remediation gigs; Fiverr said searches for "AI cleanup" services grew more than 20-fold from 2023 to 2026. SEJ is careful here and so am I — each platform measures a different thing over a different period (listings vs. gigs vs. keyword searches), so the three figures are not directly comparable.
The most common cleanup category was graphic design, followed by video editing, proofreading, and content writing. The recurring complaint from the freelancers doing it: clients budget the repair as quick and cheap, when the fix can take as much work as starting over — and some now refuse the jobs outright.
The piece includes its own counterweight rather than selling the boom. A study published in Management Science found job posts for the writing and coding work most exposed to automation fell 21% relative to less-exposed jobs within eight months of ChatGPT's launch, and image-creation posts fell 17% after image generators arrived. So the honest reading is not "AI created jobs" — it's that the work shifted from producing to repairing, and the repair market is smaller than what it replaced.
Why it matters for an independent practice: This is a price tag on the thing MMR has been arguing qualitatively for a year. AI makes the first draft nearly free and leaves the credibility entirely unpaid for — and somebody bills for that gap, either a freelancer or the practice's reputation. Two concrete reads. One, if a client is generating their own blog or social content with AI and it needs a human pass before it's publishable, that pass is a line item, not a favor — this is the market data that lets you price it honestly instead of absorbing it. Two, note where the cleanup demand concentrates: design and video first, writing fourth. The visual assets a practice thinks are the safe, low-risk thing to automate are exactly where the rework is landing.
A spine institute opened a PRP and adipose-cell center — and the press release is a clinic in how to launch regen without making a claim
Ortho Spine News · Tim Allen · September 3, 2026Regenerative medicine
Joseph Spine Institute (Tampa, FL) announced the launch of its Regenerative Therapy Center on Sep 3 via PRNewswire, offering autologous adipose-derived cell therapies and platelet-rich plasma (PRP) for spine and musculoskeletal conditions.
Read the verbs, because they're doing the work. Regenerative medicine is "an evolving field that explores the potential" of using a patient's own cells. Adipose tissue contains components "that are being studied for their potential applications." The therapies "may be considered for appropriately selected patients following a comprehensive medical evaluation." There is no efficacy claim anywhere in the release — no outcome, no percentage, no condition promised.
The founder's quote is the disqualifier, not the pitch. Samuel A. Joseph, Jr., MD: "Not every patient is a candidate for regenerative therapy, and regenerative treatment isn't appropriate for every diagnosis… We begin with a thorough evaluation to determine the source of a patient's symptoms and then discuss all appropriate treatment options. Our objective is always to match the right patient with the right treatment."
Structurally, the center is positioned as an extension of the Institute's existing spine diagnosis, interventional pain management, sports medicine and minimally invasive surgery practice — one option inside a full spectrum — rather than as a standalone regenerative offer patients come to on its own.
Why it matters for an independent practice: Every regen client MMR serves has to answer the same question this release answers: how do you announce a cash-pay biologics service without writing the sentence that draws an FDA warning letter or a state board complaint? This is the template. Claims are made about the field's promise, never about the patient's outcome; the physician's quote spends its words on who is not a candidate; and the service is framed as one option inside a diagnostic practice rather than a product being sold. That last move is the strategically interesting one — it makes the offer harder to commoditize and harder to attack, because the thing being sold is the evaluation, not the injection. Worth holding up against any client's current regen page. Most of them fail at the first verb.
Video — "OpenAI, NVIDIA And Anthropic Just Split. Here's How I'd Spend $20, $60 Or $200." (17:42, uploaded Wednesday morning)
Nate B Jones · uploaded Sep 2, 7:00 AM PDT · July 22, 2026Buildable AI
Upload date hard-verified from the page's own `uploadDate` metadata: 2026-09-02T07:00:38-07:00 — two days old, comfortably inside the days-old rule. Runtime 17:42, roughly 68K views. Channel: 319K subscribers, 1.1K videos. Linked to the specific watch URL, per the standing rule.
This is the first item this source has ever contributed. Nate B Jones was approved to Trusted on July 22, 2026 and has sat at (0/0) since — 44 days on the list without supplying an item or receiving a vote. Worth saying plainly: a Trusted source that never gets pulled from is a list-maintenance problem, and today it got pulled from.
I did not watch or transcribe this video, and I'm telling you that instead of inventing a stack, a number or a recommendation. The framing — that the major labs have diverged enough that "which subscription" is now a real allocation decision rather than a coin flip — is his, not mine, and the tiers named in the title are the only specifics I can verify from outside the video.
Why it matters for an independent practice: Every practice and every agency is now paying for AI at three or four price points across an unaudited pile of seats, and almost nobody has revisited that spend since they signed up. The allocation question is real: a $20 seat that drafts patient education, a $200 seat that a single person uses for building automations, and an API line item nobody owns are three different decisions, and the vendors diverging means the "just pick one and standardize" answer is getting worse, not better. The independence angle matters too — a practice that concentrates everything on one vendor's top tier has bought a switching cost, not a capability.
