> daily_signal(2026_06_18)

The 3D scans you fed Pokémon Go now steer robots through your city and were pitched for military drones; a quiet firmware update stripped the memory encryption from consumer AMD laptops.

PickBits Daily Signal · Thursday, June 18, 2026

By Mark Pickering · 12 min read · June 18, 2026

// tl;dr

The roughly 30 billion location scans that Pokémon Go and Ingress players submitted for in-game rewards trained a visual positioning AI that now steers a fleet of about 1,000 delivery robots and has been paired with a military contractor to guide drones where GPS does not work. Niantic spun that mapping engine off as Niantic Spatial, sold the games to Scopely, and now says current Pokémon Go data is no longer shared with it. The model that does the steering, though, was built on years of those player scans.
A silent firmware update switched off the hardware memory encryption on consumer AMD Ryzen chips, and the company will only say the feature is now reserved for its PRO line. The silicon is identical; the protection is disabled in firmware. Anyone who relied on it, a journalist, an activist, anyone with sensitive files on a laptop, lost a real defense against someone who physically steals the machine, and nobody was told.
A British detective is under criminal investigation for using AI to fabricate evidence across several cases, believed to be the first allegation of its kind in the UK. Prosecutors are now combing back through his cases to find which ones were touched. There is no rule on the books in US courts stopping the same thing, which is what makes the precedent worth watching from here.
Even Texas now wants data centers to pay for their own power. Governor Greg Abbott ordered state energy regulators to make sure residential customers do not foot the bill for data-center growth, with a directive to cut residential transmission costs by July 31. It is the clearest sign yet that the politics of who pays for the AI build-out are turning, in the reddest, most deregulation-friendly grid in the country.

Start with the version of the week's pattern that you may have helped build yourself. If you ever stood in a park and scanned a PokéStop to claim an in-game reward, you contributed to a 30-billion-image map of the real world that now guides delivery robots down city sidewalks and has been pitched to a defense firm for steering drones where GPS goes dark. That is the shape of three of today's four stories: a system you fed, bought, or stand in front of has already moved, and the switch is in someone else's hand. A firmware update quietly takes the memory encryption out of your laptop. A detective allegedly conjures evidence out of an AI and a court has to go back and find out whose cases it poisoned. The capability deploys first, and the accountability shows up late, or not at all. The fourth story is the one going the other way, and from an unlikely place: in Texas, of all states, the governor just told the power regulators that the data centers, not your household, should pay for the grid they are straining.

This week the maps players built inside a phone game turned up steering robots and military drones, a firmware update stripped the encryption out of consumer AMD laptops, a detective allegedly faked evidence with AI, and, the one lever moving back toward people, even Texas moved to make data centers pay for their own power instead of your house.

1. The maps you built playing Pokémon Go now steer robots, and were pitched for military drones.

The clearest case yet of consumer data quietly becoming infrastructure, the through-line under this whole edition.

For years, Pokémon Go and its sister game Ingress asked players to point their phones at real-world landmarks and submit short scans, in exchange for in-game items. Those scans added up: roughly 30 billion images of streets, plazas, statues, and storefronts, enough to train a visual positioning system that can pin a device's exact location and orientation from what its camera sees, down to centimeter accuracy, without GPS. Niantic, the studio behind the games, spun that mapping engine into a separate company, Niantic Spatial, led by founder John Hanke and funded with about $250 million, then sold the games themselves to Scopely for $3.5 billion. The positioning model is the asset that survived the split, and it is now in the world: it guides Coco Robotics' fleet of roughly 1,000 sidewalk delivery bots across cities including Los Angeles, Chicago, and Miami, and in December Niantic Spatial announced a partnership with the defense firm Vantor to fuse its ground-level mapping with aerial navigation for autonomous drones and military robots operating in GPS-denied environments.

Here is the honest tension, because the company contests the sharpest version of this. Niantic Spatial says that now that Pokémon Go belongs to Scopely, current player data is no longer shared with it, and that the in-game scanning feature was wound down as part of the sale. That is a fair clarification, and it means the drones are not being fed a live stream of where you played last weekend. But it does not change the load-bearing fact: the model doing the steering was trained on years of player scans submitted by people who thought they were catching Pokémon, not building dual-use navigation infrastructure. The data was gathered under one frame, a game, and is now deployed under another, commercial robotics and a military contract, and the people who supplied it were never asked about the second one.

GeekWire report 2026 Niantic Spatial unveils global 3D mapping platform built on 30 billion Pokemon Go and Ingress player scans visual positioning system centimeter accuracy powering Coco Robotics delivery bots and Vantor military drone navigation in GPS-denied environments — kotaku.com · June 2026

Why this matters: This is the cleanest example of a pattern that is going to define the next decade of AI: the most valuable training data is the stuff ordinary people generate without thinking of it as data at all. A game told millions of players that scanning a landmark was worth a free item, and that exchange quietly built a 3D model of the physical world precise enough to drive a robot and useful enough to interest the military. The structural gap is consent: there is no mechanism, and no law that requires one, for the people who generated a dataset under one purpose to have any say when it is repurposed for another. The lever that exists is narrow but real, the data-deletion and opt-out rights that privacy laws in a handful of states now grant. Action this week: If you played Pokémon Go or Ingress, you can ask Niantic to delete your data through its privacy portal (nianticlabs.com/privacy), which takes about ten minutes and is the one direct say you have over scans you already submitted. More broadly, before you accept the next "scan this for a reward" prompt in any app, treat it as what it is, you are donating a permanent, repurposable asset for a one-time perk, and decide if the trade is worth it. If you build products, the takeaway is sharper: the purpose you collect data under is not the purpose you will be tempted to use it for later, and writing that boundary into your own terms now is cheaper than defending the absence of it.

geekwire.com: From Pokémon GO to physical AI, Niantic Spatial unveils its global 3D mapping platform (June 2026)
dronexl.co: Pokémon Go scans quietly trained the navigation tech now headed into military drones (June 9, 2026)
kotaku.com: Pokémon Go data wasn't shared with drone company, Niantic Spatial says (June 2026)

2. A firmware update quietly switched off your AMD laptop's memory encryption.

The same shape as story one, a change you did not choose, pushed to hardware you already own.

If you bought a computer with a recent AMD Ryzen chip, it likely shipped with a security feature called TSME, transparent secure memory encryption, that scrambles everything held in RAM using a hardware key that is regenerated every time the machine boots. It runs invisibly and it defeats a specific, real class of attack: someone who physically gets your laptop and tries to read its memory directly, by freezing the RAM chips, snooping the memory bus, or pulling the modules out to read elsewhere. Then a firmware update, delivered through the motherboard's AGESA microcode, silently turned it off on consumer Ryzen parts. The change was spotted in April by a privacy-focused Linux user, Ben Kilpatrick, installing a fresh OS on a Ryzen 7 9700X, and confirmed since: the feature now works only on AMD's PRO business chips. The catch is that the silicon is identical. The consumer chip is fully capable of encrypting memory; the firmware simply tells it not to.

AMD has said almost nothing. Pressed by reporters, the company offered a single line, that TSME "is a security feature only applied to PRO CPUs as part of AMD PRO Technologies," and declined to explain when the change happened, why, or why it went out with no notice. That silence is the part worth sitting with. Full-disk encryption like BitLocker or FileVault still protects your files when the machine is fully powered down, so this is not a reason to panic about a stolen-and-shut-off laptop. What you lost is the layer that protected memory while the machine was running or asleep, which is exactly the state a laptop is in when it is grabbed off a café table or seized at a border. For most people that is an abstract risk; for the journalists, activists, lawyers, and researchers who specifically chose AMD partly for that protection, it is a concrete one, removed without their knowledge.

Tom's Hardware report June 2026 AMD silently removes TSME memory encryption from consumer Ryzen CPUs via AGESA firmware update leaving users unaware they may be vulnerable feature now reserved for PRO chips identical silicon disabled in firmware — tomshardware.com · June 16, 2026

Why this matters: A protection you paid for and may have chosen on purpose was disabled remotely, with no announcement and no opt-out, and the only way most owners would ever find out is a security researcher noticing by accident. That is the same loss of control as the Pokémon Go story from the other direction: there, your data moved without you; here, your device's defenses moved without you. The structural problem is that "security feature" and "product segmentation" are now the same lever, the same chip can have a protection or not depending on which market the vendor wants to charge, and the firmware that decides is updated silently. Action this week: Check whether your machine has a Ryzen PRO or a consumer Ryzen chip (on Windows, open System Information and read the processor name; "PRO" will be in it if you have it). If you are on a consumer chip and you handle sensitive material, make sure full-disk encryption is on (BitLocker on Windows, which takes a few minutes in Settings under Device Encryption), and change your habit of leaving the laptop asleep, fully shut it down when you leave it unattended in public, because shut-down is the state where disk encryption actually protects you and the lost TSME layer does not. If you are a security or IT lead, audit which of your fleet relied on TSME and decide whether the threat model now requires PRO hardware or a software mitigation.

tomshardware.com: AMD silently removes memory encryption from consumer Ryzen CPUs (June 16, 2026)
thenextweb.com: Your Ryzen CPU used to encrypt your RAM. A firmware update silently turned that off (June 2026)
slashdot.org: Users cry foul after AMD stripped memory crypto from its consumer CPUs (June 15, 2026)

PickBits Daily Signal is free. If this lands in your inbox every day and it is worth something to you, the best way to support it is to share it with someone who would read it. Subscribe at pickbitsai.substack.com.

3. A detective is the first UK officer accused of faking evidence with AI.

The accountability beat, where the system that already moved is the criminal justice system itself.

A serving officer with Derbyshire Constabulary in England is under criminal investigation over allegations that he used artificial intelligence to fabricate evidence across several cases, in what is believed to be the first allegation of its kind against a police officer in Britain. The suspected offense is perverting the course of justice, one of the most serious charges in the system because it strikes at whether a trial can be trusted at all. No arrests have been made yet, and the officer has not been named. What makes the story bigger than one bad actor is the cleanup: the Crown Prosecution Service is now working with the force, defense teams, and the courts to identify which cases were affected, meaning every prosecution the officer touched is now in question, including any that already ended in a conviction.

This is not the first warning. Last year, West Midlands police relied on AI-generated material that fabricated a match in a case involving the football club Maccabi Tel Aviv, and forces on both sides of the Atlantic have been scrambling to write rules for AI tools faster than officers adopt them. The reason it should land for a US reader is the absence on this side: there is no settled rule in American courts squarely barring AI-fabricated exhibits, and the federal judiciary is only now debating how the rules of evidence should handle deepfakes and synthetic media. Generative tools can now produce a photo, a document, or an audio clip convincing enough to enter a case file, and the institutions that decide guilt and innocence are running on authentication procedures written for an era when faking that material was hard. The Derbyshire case is the first time a court has had to ask, in public, how many real cases a single person with these tools could have quietly corrupted.

ITV News report June 14 2026 Derbyshire police officer investigated over alleged use of AI to create evidence first case of its kind in Britain perverting the course of justice Crown Prosecution Service identifying affected cases — cybernews.com · June 14, 2026

Why this matters: The thing that is supposed to be hardest to fake, the evidence that can take away someone's freedom, just got easier to fake, and the first known instance is an insider with a badge rather than a hacker. If you are ever a juror, a defendant, or a witness, the quiet implication is that a photo or document entered as fact may now need to be authenticated in a way courts have never routinely required. The structural gap is procedural: rules of evidence assume forgery is expensive and detectable, and generative AI breaks both assumptions, so until courts adopt provenance standards, the safeguard is whoever in the room thinks to question a file's origin. Action this week: If you work anywhere near the legal system, as a paralegal, an investigator, an expert witness, an insurance adjuster, read up on content-provenance standards like C2PA (contentauthenticity.org), the emerging way to cryptographically sign where a piece of media came from, because demanding provenance is fast becoming the practical defense against synthetic evidence. For everyone else, the durable habit is simpler: a screenshot, a recording, or a document is no longer self-proving, and "I saw it with my own eyes" now has to include "and I checked where it came from."

itv.com: Derbyshire police officer investigated over alleged use of AI to 'create evidence' (June 14, 2026)
cybernews.com: UK police officer investigated for using AI to create evidence in cases (June 2026)
computing.co.uk: Derbyshire police officer investigated over alleged AI-generated evidence (June 2026)

4. Even Texas now wants data centers to pay for their own power, not your house.

The constructive counterpoint, and the one lever this week that moved back toward ordinary ratepayers.

On June 10, Texas Governor Greg Abbott issued a directive to the state's two energy authorities, the Public Utility Commission and the grid operator ERCOT, with a blunt premise: "everyday Texans" should not be "burdened with the costs of infrastructure driven by data center expansion." He told the PUC to require data centers to pay the full cost of the power infrastructure built for them, so residential customers bear none of it, and to initiate action to cut residential transmission costs by July 31. He gave both agencies until July 17 to report back on what else they can do under existing authority. And he laid out six policies for the legislature's 2027 session, including repealing the sales-tax breaks that lured data centers to Texas in the first place, requiring new facilities to add their own generation capacity, mandating water-efficient cooling, and setting noise and community standards.

The significance is in who is saying it. Texas runs the most deregulation-friendly grid in the country and has spent years rolling out the welcome mat for data centers; for that governor to call for making them pay their own way, and to put repealing his own state's incentives on the table, is a real turn in the politics of the AI build-out, not a blue-state talking point. The honest caveat is that this is a directive and a wish list, not a law yet: the regulators still have to act, the legislature does not convene until 2027, and data-center developers are already pushing back, with one, Fermi, publicly responding within two days. But the direction is unambiguous, and it matters far beyond Texas, because the question Abbott is forcing, should the people who profit from a data center pay for the grid it strains, or should the household down the road, is the exact fight now playing out in front of utility regulators in nearly every state racing to host AI.

Texas Tribune report June 10 2026 Governor Greg Abbott recommends sweeping data center regulation directs PUC and ERCOT to shield residential ratepayers from data center costs cut residential transmission costs by July 31 repeal sales tax exemptions — texastribune.org · June 10, 2026

Why this matters: The single most direct way the AI boom reaches your life is your electric bill, because data centers draw enormous power and someone has to pay to build the lines and plants that feed them. For most of the last two years the default answer, in state after state, has been that the cost gets spread across all ratepayers, meaning your household helps subsidize a hyperscaler's power. Abbott just put the opposite principle on the record in the country's biggest energy state: the data center pays for the data center. The structural lever here is your state's public utility commission, the unglamorous body that decides how grid costs get split, and it is where this fight is actually won or lost. Action this week: Find out whether your state's utility regulator has an open docket on "large load" or data-center cost allocation, most now do, by searching your state's public utility commission site, and if there is a public comment window, a few sentences from a resident saying you should not subsidize data-center power lines is exactly the input those proceedings weigh. If you want the national frame, the Federal Energy Regulatory Commission is running its own large-load interconnection rulemaking (ferc.gov), which will shape the rules even where your state lags. And if you are simply shopping for where to live or expand a business, the cost of who-pays-for-data-centers is about to become a real line item in your power rates.

texastribune.org: Greg Abbott recommends data center industry crackdown (June 10, 2026)
gov.texas.gov: Governor Abbott directs PUC and ERCOT to shield Texans from data center infrastructure costs (June 10, 2026)
newschannel10.com: Fermi responds to Gov. Abbott's data center directive (June 12, 2026)

» What to watch this week

Whether any procurement record surfaces showing a robotics or defense buyer licensing Niantic Spatial's positioning model, and whether the Vantor military partnership produces a named program. The tell is a contract or a deployment, which would move "pitched for drones" to "flying on drones," and whether any state's privacy regulator treats repurposed game-scan data as a consent question.
Whether AMD reverses the TSME removal or clarifies it under pressure, and whether a class of buyers, enterprise or government, forces the issue by refusing consumer chips that lack it. Watch for a firmware update that quietly restores the feature, which would be the cleanest admission that pulling it was a mistake.
Whether the Derbyshire investigation produces charges and how many convictions the CPS review unwinds, and whether a US court or rules committee responds with an actual standard for authenticating digital evidence. The next data point is either a charging decision in England or a Federal Rules of Evidence proposal here.
Whether the Texas PUC actually moves by its July 17 and July 31 deadlines, and whether other states' regulators follow with their own data-center cost-allocation orders. A directive that turns into a binding PUC order is a trend; a directive that stalls before the 2027 legislature is a headline.

Tomorrow's signal lands here.