> daily_signal(2026_08_27)
San Francisco banned the software that sets your rent nearly two years ago and has fined no one, the same week AI got a hospital billing code and New York warned it is emptying people's savings.
PickBits Daily Signal · Wednesday, August 27, 2026
// tl;dr
- San Francisco banned landlords from using RealPage-style rent-setting software in 2024, but nearly two years on it has brought zero enforcement actions while asking rents rose about 14% in four months. The first real challenge is a tenant's federal suit, Gomez v. Greystar, not a city case, as more than a dozen cities and states pass their own bans on the practice New Jersey's governor calls "collusion by algorithm."
- The AMA is advancing billing codes that would let hospitals charge Medicare for AI that reads scans, labs, and cardiac rhythms with no physician at the point of care. The codes ride a new "Software as a Medical Service" category in CMS's 2027 rule, the comment window closes August 31, and nurses, about 30% of hospital labor cost, still have no code of their own.
- New York warned residents on August 26 that AI-cloned voices, deepfake video, and fake celebrity endorsements have made investment scams look legitimate. The FTC counted 144,041 people reporting more than $8 billion in investment-scam losses in 2025, up 38% year over year, median loss $10,560; the tells that still catch them have not changed.
- Engineers at the FAMU-FSU College of Engineering built GridFusionX, an AI that forecasts electricity demand and renewable generation 56% more accurately in tests. It cut the standby reserve power utilities must hold by up to two-thirds, closing the week's grid-strain thread on a rare up-note. It is published research, not a deployed product.
Rent is the one that got me this week. San Francisco banned the RealPage software nearly two years ago, the stuff that sets your rent off your neighbors' private numbers, and in that time it has fined exactly no one while a tenant does the city's job in federal court. We have watched the surveillance-pricing version of this for months, stores charging you and the person next to you different prices off the same data. Rent is just the version that hits everyone.
The hospital one is worse in a quiet way. We watched an AI out-diagnose a six-doctor panel back in June, and now the AMA is writing the billing code that pays for that kind of read, while the nurse in the room still has no code of her own, something nurses say they've wanted for a century. New York's scam warning isn't new either; it's the voice-clone story we picked up when it was draining older people's retirement accounts, except now there's an $8 billion number on it. The last one is the good one, a Florida lab's grid AI that for once makes the system cheaper instead of pricier. In all four, the technology is already running while the rules are still catching up, and that gap is the thing I keep coming back to.
San Francisco wrote the rent ban, RealPage kept setting the rent, and the only party who took it to court was a tenant.
1. San Francisco banned the software that sets your rent nearly two years ago.
Nearly two years on the city attorney has not filed a single enforcement action, while asking rents jumped about 14% in four months to record highs.
Start with the document. In October 2024, San Francisco's ordinance, Admin. Code § 37.10C, made it illegal for landlords to use revenue-management software like RealPage and Yardi that recommends rents from nonpublic competitor data, with civil penalties up to $1,000 a violation. The software is genuinely useful to a landlord, and almost nobody disputes that it works; the fight is about what it does with everyone's private numbers at once. New Jersey Gov. Mikie Sherrill, who signed her state's FAIR Act in July making the practice an antitrust violation the attorney general can enforce, calls it "collusion by algorithm."
Then nothing happened. Nearly two years after the ban took effect, the city attorney has filed zero enforcement actions, and over that stretch San Francisco asking rents rose about 14% from March to July 2026 to record highs (ApartmentList). Ordinance author Aaron Peskin concedes his market-cure expectation predated "a trillion dollars of AI capital." The first real test came from a tenant, not the city: Gomez v. Greystar was filed in the Northern District of California in July, one of a wave of suits as more than a dozen jurisdictions, from Seattle (up to $7,500 a violation) to Philadelphia, enacted their own bans. That many places passing the same law in a year tells you the problem is real; whether any of them actually enforces it is the open question.
Why this matters: There is a real chance the rent you pay was set by software reading your neighbors' leases, and a law that supposedly banned it where you live may be doing nothing about it. A ban no one enforces isn't protection, it's a press release, and San Francisco is the proof: nearly two years, zero fines. The teeth so far are elsewhere: the private antitrust suits, and New Jersey's AG-enforceable FAIR Act.
Action this week: Find out whether your building prices with RealPage or Yardi, which usually shows up as identical, algorithm-timed increases across one landlord's units, and report suspected use through your city or state tenant portal; New Jersey stood up a reporting site for exactly this. When I have watched these cases, the ones that actually move are the tenant suits, not the city ordinances, so if your rent climbed in lockstep with the building next door, that pattern is the thing worth documenting now.
sfstandard.com: SF banned rent-setting software in the AI boom, and hasn't enforced it (August 21, 2026)
morganlewis.com: Algorithmic rent-pricing litigation expands under new state and local laws (August 2026)
governing.com: New Jersey bans AI-driven rent-setting (2026)
2. The AMA is writing billing codes so hospitals can charge for AI that reads your scans with no doctor in the loop.
The codes ride a new Medicare "Software as a Medical Service" category, while bedside nursing, about 30% of hospital labor cost, still has no code of its own.
If you sit anywhere near a hospital's revenue cycle, there is a comment window that closes August 31 and a decision behind it you will live with. The AMA's CPT Editorial Panel is advancing a framework it calls Clinically Meaningful Algorithmic Analyses, which would let a hospital bill for AI-generated reads, diagnostic images, lab panels, and cardiac rhythms, with no physician required at the point of care. It ties into a new Software as a Medical Service category in CMS's proposed 2027 Medicare outpatient rule.
The term that should stop you is on the liability line. CMAA billing recognition runs on a separate regulatory track from FDA clearance, so who answers for an inaccurate autonomous read is, right now, undefined. And the reimbursement itself is the tell for where this goes. Bedside nursing, about 30% of hospital labor cost, still has no direct billing code and stays bundled into the daily room charge. Rebecca Love, RN, who founded the Commission for Nurse Reimbursement, put it plainly: if AI shows up as a billable line item and nursing stays buried in the room charge, the money will follow what is visible on the ledger. We watched an AI out-diagnose a six-doctor panel in June; this is the billing plumbing that decides who gets paid when it does.
Why this matters: The next time an AI reads your scan, a payment system is deciding that read deserves its own line on the bill before the nurse at your bedside gets one. A billing code is not paperwork; it is how a hospital decides what work is worth paying for, and once the AI read is on that list, the budget and the staffing follow it. And nobody has figured out who is liable when the AI gets a read wrong, which is the thing to pin down now, while the rule is still open.
Action this week: Comment to CMS on the 2027 outpatient rule at regulations.gov before August 31, and put one question on the record: who carries the liability when an autonomous read is wrong, and in which document. Every time I have watched one of these blow up, it turned out nobody wrote down who was responsible before the money moved. If you own a revenue cycle, get the FDA-clearance-versus-billing distinction from your vendor in writing while the comment window still gives you leverage.
natlawreview.com: The AMA's new AI billing codes draw nursing opposition (August 19, 2026)
nurse.org: AI billing CMAA CPT codes and nursing reimbursement (August 2026)
3. New York warned residents this week that AI has made investment scams almost impossible to spot.
The FTC counted more than $8 billion in investment-scam losses last year, up 38%, yet the tells that still catch them have not changed.
On August 26, New York's Division of Consumer Protection and Department of Financial Services jointly told residents that AI now makes investment scams look legitimate, using cloned voices, deepfake video, and fake celebrity endorsements. Secretary of State Walter T. Mosley offered the oldest rule there is, "if it seems too good to be true, it probably is." A state can issue an alert. What it cannot do is put friction on the platforms where the fakes spread.
The numbers are the reason for the alarm. The FTC logged 144,041 people reporting more than $8 billion in investment-scam losses in 2025, up 38% in a year and its single costliest fraud category, with a median loss of $10,560. Separately the FBI's 2025 report counted $893 million in AI-enabled-crime losses across 22,364 complaints. We have been watching this one for months, back when AI voice clones were already draining elderly retirement accounts. The fakes got better; the tells stayed exactly the same. Guaranteed returns, pressure to act fast, and a push to move the conversation to a private messaging app are still the signature.
Why this matters: A voice you would recognize anywhere can now be faked well enough to talk you, or your parents, out of your savings. The surge is national, not a New York problem, and the median hit is over ten thousand dollars. What has not moved is where you report it and how you verify, which is the one advantage left to the rest of us.
Action this week: Set a family safe word with your parents and anyone older in your life, because the AI-cloned "grandkid in trouble" call is exactly what this money funds. Verify any adviser or product through the SEC before you send a dollar, and if you are hit, report it to the FTC, the FBI's IC3, and your state attorney general the same day. My own rule is a callback on a number I already have, never the one that just called me, and New York's consumer hotline, 1-800-697-1220, is a good place to start.
fingerlakes1.com: New York warns consumers as AI investment scams surge (August 26, 2026)
malwarebytes.com: Americans lost nearly $900 million to AI-powered scams, FBI says (June 2026)
PickBits Daily Signal is free, and it is the fastest daily read on where AI's real costs and capabilities are landing. If it is worth something to you, the best support is to forward it to one person who would read it. Subscribe today!
4. Florida engineers built an AI that forecasts the power grid 56% more accurately.
In tests it cut the costly backup power utilities burn to stay reliable by up to two-thirds, a rare bit of good news in a month of grid-strain stories.
End on the good one. Picture the grid operator watching a screen, trying to keep the lights on without paying to hold a mountain of standby power for a spike that may never come. That's who this is for. Researchers at the FAMU-FSU College of Engineering and FSU's Center for Advanced Power Systems built GridFusionX, an AI that uses graph neural networks to forecast electricity demand and renewable generation across interconnected regions, and returns its predictions with confidence intervals so an operator knows how much to trust them.
In real-world tests it improved forecasting accuracy by up to 56% and cut the reserve power utilities must hold in standby by up to 66%, which the team links to steadier reliability, better wind and solar integration, and more accurate consumer bills. The stage matters here: this is a published research result in IEEE Transactions on Network Science and Engineering, led by doctoral student Quoc Bao Phan, not a tool any utility has deployed. All week we have covered the other direction, data centers loading the grid and pushing your bill up. This is a lab aiming the same tools at forecasting it better.
Why this matters: The AI blamed all year for straining the grid and lifting your power bill can also forecast that grid well enough to waste less of it. Better forecasting is exactly what lets a system lean on wind and solar without hoarding expensive standby power, so this is not only a story about AI loading the grid. It is a method proven on test data, though, not a product on a utility's floor.
Action this week: Watch for the first utility or grid operator to run a pilot on graph-neural-network forecasting against its own reserve-margin practice, because the up-to-66% reserve cut is the number to see reproduced outside a paper. I keep a short list of results like this that never leave the journal, and the thing that would move GridFusionX off it is a named ISO or utility putting it on real load. Until then, treat it as the better-forecasting-method story it is, not a lower bill you can expect next month.
» What to watch this week
- Whether San Francisco's city attorney files a single RealPage enforcement action, or the tenant suits keep doing the city's job. Nearly two years of a ban with zero fines is the tell; the open question is whether Gomez v. Greystar and New Jersey's FAIR Act make enforcement the norm or leave it the exception.
- Whether CMS advances the CMAA billing pathway in the 2027 rule, and whether the liability question gets answered before it does. The comment window closes August 31; watch whether anyone names who is accountable for a wrong autonomous read, and whether nursing's own code gets any traction in the same cycle.
- Whether other states follow New York with formal AI-scam warnings, and whether any friction ever lands on the platforms hosting the deepfakes. An alert tells residents to be careful; the real move is regulators going upstream to where the fakes actually spread.
- Whether a real utility or ISO runs a pilot on GridFusionX-style forecasting. A 56% accuracy gain on test data proves out only when a system operator puts it on live load and the reserve savings show up in practice, not just in the journal.
Tomorrow's signal lands here.