> daily_signal(2026_11_16)
New York City banned surveillance pricing, a central bank put data behind AI thinning young workers' jobs, records showed a Medicare AI built to deny care, and a lab read genes off an ordinary slide.
PickBits Daily Signal · Monday, November 16, 2026
// tl;dr
- The New York City Council voted 44-5 to ban "surveillance pricing" - stores using your personal data to set an individual price for the same product. It covers about 8 million consumers, with fines up to $1,000 per violation enforced by the Department of Consumer and Worker Protection. NYC is the second US city after Seattle; New York State's own June bill still sits unsigned, and in October the FTC ordered Uber, DoorDash, Walmart and Target to explain how they price.
- A Reserve Bank of Australia analysis found AI is thinning jobs for the youngest workers first. Job postings have fallen most in the highest-exposure occupations - programming, clerical work, customer service - and workers aged 15-24 now show weaker employment than older peers. US payroll data backs it: Stanford found a 13% relative drop for 22-to-25-year-olds in exposed jobs, Anthropic a ~14% fall in their job-finding rate.
- Records EFF obtained through a FOIA lawsuit show the AI prior-authorization program inside traditional Medicare was built to deny. CMS picked six vendors from 66 for its WISeR pilot and has sent corrective notices to at least four; one vendor denied more requests than it approved, one runs on Palantir infrastructure, and the payment structure itself gives vendors an incentive to deny as many claims as possible. The pilot runs through 2031.
- Researchers introduced GeneFlow, an AI that predicts spatially-resolved gene activity directly from a routine, cheap tissue slide. It reports up to 56.8% gains on 1,000-gene prediction and cuts the compute cost by orders of magnitude, so any lab with a microscope can now run something that used to need an expensive assay. It is benchmark-stage, not a cleared clinical test.
Start with the New York City Council. It voted 44-5 this week to stop stores from using your own data to decide what you pay, fines up to a thousand dollars a time. The same week, Australia's central bank ran the numbers on AI and young workers and found the bottom rung of the job ladder thinning first: programming, clerical work, customer service. And the latest batch of EFF's FOIA records showed the AI now screening some Medicare claims was set up to pay its vendors more when they say no. I've followed all three for months; what's new this week is just that we finally have real numbers on all three.
I'm saving the good one for last. A team of researchers got an AI to read gene activity off a plain tissue slide, the kind your hospital already makes, for a fraction of what the usual lab test costs. For once it's pointed at a problem instead of at us.
The best thing AI did this week was read a tissue slide instead of a shopper.
1. New York City voted 44-5 to ban stores from using your personal data to set your price.
It's the second US city to do it after Seattle, and more states are lining up behind it.
On November 12, 2026, the New York City Council went 44-5 to bar businesses from using your personal data - browsing, location, purchase history - to charge different people different prices for the same thing. Advocates call it surveillance pricing. The city's Department of Consumer and Worker Protection enforces it, with fines up to $1,000 per violation and an exception for ordinary, transparent discounts. Consumer Reports' Grace Gedye put the reach at about 8 million NYC shoppers.
Stores have done this quietly for years. What changed this week is a city this big making it illegal, and New York isn't the only one moving. NYC is the second US city to act after Seattle. New Jersey, Connecticut and Maryland already have bans. New York State passed its own version in June that Governor Kathy Hochul has still not signed. And in October the FTC ordered nine companies, among them Uber, DoorDash, Walmart and Target, to explain how they set prices. Now watch which of those FTC inquiries turns into real enforcement, and whether Albany signs its own bill.
Why this matters: A city just made it illegal for a store to use your personal data to show you a different price than the shopper standing next to you. This one doesn't make headlines like the chatbots do, but it hits you every time you buy something. New York wrote the fines in a way another city can copy almost word for word, and I expect a few will.
Action this week: Check what actually feeds your prices, and flag anything tied to a specific shopper: where they are, their device, what they've browsed or bought. A normal published discount is fine; a price that quietly changes because of who's looking is the thing that now costs $1,000 a pop. My read is every big US market is heading this way, not just New York - we watched a wave of these bills move through statehouses earlier this year, and this is the biggest one to actually pass. If you're a shopper, the place you've actually got a say is local - the DCWP complaint line is how you make them act.
2. Australia's central bank found AI is thinning jobs for the youngest workers first.
US payroll data from Stanford and Anthropic points the same way: the squeeze early-career workers have felt now shows up in the numbers.
Last month we covered the Gallup survey where employers said a college degree matters more in the AI age. A Reserve Bank of Australia analysis by Mike Major, "Monitoring early effects of AI on the labor market," reported November 15, puts hard numbers on the other side of it. Since ChatGPT launched in late 2022, job postings have fallen across the board. The steepest drop is in the most AI-exposed work - programming, clerical, customer service. And inside that group, workers aged 15-24 and recent graduates now show weaker employment and higher unemployment than older people doing the same jobs.
This is a central bank talking, not an advocacy group, and they're upfront that it's correlation, not proven cause. Two US datasets land the same way. Stanford's August 2025 payroll study found employment among 22-to-25-year-olds in AI-exposed jobs fell 13% in relative terms, nearly 20% among young software developers; an Anthropic study in March 2026 found their job-finding rate fell about 14% with no systematic rise in overall unemployment. Three independent datasets, same direction - that's the part I can't wave off.
Why this matters: If you manage a team or you're early in your career, the bottom rung is thinning, and now a central bank has the numbers. Junior developer, first-line support - the jobs people used to start in are the ones the exposure data shows shrinking fastest.
Action this week: Decide whether you're still training your juniors or quietly letting AI absorb them, because otherwise you look up one day and there's no one left to train. I covered the Gallup survey last month, where employers said a degree matters more now. The RBA finding is the flip side of it. Keep the junior roles where a person actually runs the AI, and be honest about which ones AI just does now. Share it with whoever owns hiring on your team.
Listen to the Daily Signal podcast at pickbits.ai/daily-signal.
3. Records EFF pried loose show the AI screening Medicare claims was built to deny care.
One vendor denied more requests than it approved, and CMS's own payment structure, the files show, rewards vendors for saying no.
Since January 1, 2026, a federal pilot called the WISeR Model - Wasteful and Inappropriate Service Reduction - has required doctors in six states to get AI-backed prior authorization before delivering some services in traditional Medicare. That's the version that never used to put people through this. On October 27, the Electronic Frontier Foundation published a second batch of CMS records, more than 800 pages pried loose through a Freedom of Information Act lawsuit, and they show the experiment going as critics feared. CMS picked six vendors from 66 applicants and has already sent requests-for-information or corrective-action plans to at least four of them.
One vendor denied more requests than it approved in the program's first months. One, Zyter, runs its denials on Palantir software infrastructure. Innovaccer got a non-compliance notice in February; Virtix was cited in March for turnaround times that "exceed established guidelines." And CMS's own payment structure, the records show, gives vendors "an incentive to deny as many claims as possible" - the clearest line in the file, and the whole problem right there. The demonstration is scheduled to run through 2031, and EFF is still receiving records.
Why this matters: The government is letting an AI help decide whether Medicare pays for your care, and newly released records show the program was built in a way that pays its vendors to say no. Traditional Medicare was the version that didn't do this to people; the pilot brings private insurance's deny-to-save-money reflex into the public program, and it lands on old people on Medicare, who can't easily fight back when a claim gets denied.
Action this week: Ask your provider whether a procedure routes through WISeR if you or someone you care for is on traditional Medicare in one of the six pilot states, keep a dated record of every request and denial, and appeal - an initial AI denial can be overturned on human review. This isn't new ground for us: we've followed EFF's FOIA fight over these records for months, and every document drop since has made the same program look worse each time. If this is your family's situation, send it to whoever handles their care decisions.
4. An AI now reads gene activity straight from the routine tissue slide your hospital already makes.
GeneFlow reports up to 56.8% gains on 1,000-gene prediction while cutting the compute bill by orders of magnitude - a costly lab test turned cheap.
There's a lab technique called spatial transcriptomics that maps which genes are switched on in each spot of a tissue sample. Researchers use it to see how a tumor's cells differ across a single sample, and it's brutally expensive, which keeps it out of reach for most labs and clinics. In a paper posted November 13 and accepted to KDD 2027, researchers led by Songming Zhang introduced GeneFlow, an AI that predicts that same spatially-resolved gene-expression map directly from a routine, cheap histology image - the kind of stained slide already sitting in every pathology lab.
It reports up to 56.8% gains on 1,000-gene prediction and state-of-the-art accuracy across benchmarks while, the authors say, cutting the computational cost by orders of magnitude. It's still early, though, benchmark-stage, not a cleared clinical test.
Why this matters: This AI reads gene activity off the same kind of tissue slide your hospital already makes, and it could make cancer and disease research far cheaper. I think it matters more than most of the week's chatbot noise: make an expensive test cheap, and the small labs and clinics that could never afford it finally get it too.
Action this week: Before you pay for another spatial-transcriptomics run, test whether a model reading a plain slide can match the assay on your own tissue - GeneFlow is open on arXiv now, and it's still benchmark-stage, not a cleared diagnostic, so treat it as a candidate to test rather than a result to trust. My own read is this gets a fraction of the attention the week's chatbot headlines do, and it shouldn't.
» What to watch this week
- Whether Governor Hochul signs or weakens New York State's own surveillance-pricing bill when the session resumes, and whether the FTC's October order to Uber, DoorDash, Walmart and Target turns into enforcement. The state-level bills we tracked all spring are starting to pass, so watch whether lobbyists shrink what counts as "personal data" before more cities lock it in.
- The RBA's "exposure, not causation" caveat gets tested every time new payroll data drops, and what matters is whether employers who say a degree still counts keep hiring juniors. Talk is cheap here; watch what firms actually do over the next few hiring cycles.
- The next batch of WISeR records - EFF is still pulling them out of CMS, and more is coming. The pilot runs through 2031, so watch whether CMS actually rewrites the payment structure that rewards denials, or just keeps sending corrective notices while it runs.
- Whether GeneFlow gets independent validation on real clinical tissue, not just benchmarks. Most of these never make it out of the lab; the jump from a good benchmark to a cleared diagnostic is where they stall, so watch for anyone reproducing it outside the authors' own test set.
Tomorrow's signal lands here.