Anthropic says small AI models could start copying themselves between poorly secured devices by the end of 2026 — and clinics are full of poorly secured devices
October 15, 2026 · 5 items
Anthropic says small AI models could start copying themselves between poorly secured devices by the end of 2026 — and clinics are full of poorly secured devices
Medical Economics · October 13, 2026Practice operations
In a report published Oct. 12, Anthropic's Frontier Red Team found that compact, openly available models can already complete many of the individual steps needed to self-replicate: scanning a network for targets, copying the model to a new machine, and getting it running on unfamiliar hardware.
To show cheap hardware can host a capable model, the team bought 15 off-brand Android TV boxes at about $55 each, networked them, and split a 26 billion-parameter model across them — generating 0.39 tokens per second.
The researchers did not test medical devices; the experiments used consumer electronics and emulated Android systems. But Medical Economics' Todd Shryock notes the weaknesses described are the familiar ones in clinical settings: default credentials, outdated software, and devices that rarely receive security updates.
The framing matters for practices that run networked equipment — imaging, monitors, check-in tablets — on the same flat network as everything else.
Why it matters for an independent practice: This is an inventory task, not a philosophy debate: pull the list of every internet-connected device in the office, confirm none still runs factory default credentials, and ask the vendor in writing when each last received a firmware update. That's a one-afternoon action with a dated research report behind it.
Function and Google Health are bundling 160+ lab tests, wearable signals and an AI coach into a $365-a-year consumer product
Longevity.Technology · October 13, 2026Regenerative medicine
The partnership will let US users share Function lab results into the Google Health app later this fall, combining wearable signals — resting heart rate, sleep patterns, heart rate variability, cardio load — with Function's lab panel for context.
Function claims members get 160+ lab tests including biannual testing for an annual fee of $365, with MRIs and CTs available at additional cost. The targets named are cardiovascular and metabolic markers plus advanced hormone and nutrient testing.
Google Health Coach will use personal health records to tailor guidance — and Longevity.Technology's own write-up carries the caveat plainly: AI-generated coaching may be inaccurate or incomplete.
Reported from the companies' PR Newswire announcement; this is a partnership announcement, not published outcomes data.
Why it matters for an independent practice: If a longevity or regen-med practice sells 'advanced panel + personalized plan' as the front door, that offer is now being priced at $365/year by two very large brands. The defensible ground is the part Google explicitly disclaims — a physician who interprets the panel, owns the follow-up, and is accountable for it. Price and panel breadth are the wrong things to compete on this quarter.
Amex is publishing a framework so AI agents can only buy what the user actually authorized
TechCrunch · October 13, 2026Buildable AI
American Express is introducing a framework designed to ensure AI agents only make purchases their users have authorized, which TechCrunch frames as a way to prevent costly mistakes and disputes as AI-powered shopping grows.
The stated reason is structural, not hypothetical: personal AI agents are increasingly transacting on their users' behalf, and the nondeterministic nature of large language models makes bots prone to errors.
Reported by Sarah Perez; the piece is an announcement of the framework rather than a study of agent error rates.
Why it matters for an independent practice: The same authorization question arrives in a practice the moment an agent touches anything with consequences — booking, refunds, ad spend, supply reorders. Borrow the pattern before you need it: write down, per agent, exactly what it may execute without a human approving, and what it may only draft. A card network formalizing this is the clearest signal yet that 'the agent did it' is not going to be an acceptable answer.
Floqer is selling AI agents that continuously research prospects — and plans to open-source the database behind them
TechCrunch · October 14, 2026Buildable AI
Floqer describes itself as an autonomous customer knowledge platform: an AI-powered database that continuously collects, organizes, and updates the information B2B sales and marketing teams use to understand their customers.
The startup told TechCrunch it plans to open-source the technology behind its autonomous customer knowledge database, targeting early 2027 pending security reviews.
It says it has surpassed $1.2 million in annual recurring revenue, with overall revenue up roughly tenfold over the past year. Founded by CEO Shivam Mahajan and Zaaheda Islam; pitching Wednesday on the Startup Battlefield stage at TechCrunch Disrupt 2026.
Why it matters for an independent practice: The buildable shape here isn't 'buy Floqer' — it's the continuously-refreshed account record. For a practice doing referral development, the same pattern applies to referring physicians and groups: an agent that keeps a living file on each referral source (new providers, new locations, recent moves) so outreach is specific instead of generic. The open-source commitment is worth a calendar note for early 2027.
Gudea is pitching itself as a 'storm tracker for the internet' that spots reputation damage before it erupts
TechCrunch · October 14, 2026Patient acquisition
Gudea tracks how stories and rumors spread online, and describes itself as the 'storm tracker for the internet.'
TechCrunch's framing names the threat categories directly: a seemingly insignificant post — AI-generated misinformation, hate speech, or a damaging rumor — can quickly snowball into a much larger public controversy.
Reported by Lauren Forristal; the piece is a company profile rather than an evaluation of the product's detection accuracy.
Why it matters for an independent practice: For a physician, reputation is the asset, and the new failure mode is an AI-generated claim about you that no one on staff sees until a patient repeats it in the exam room. Even without buying anything, the operational takeaway is to run a standing monthly check on what's being said about the practice name and each provider's name — in search, in reviews, and in the AI assistants patients are now asking.
