Sign in Subscribe
Century Report

Supreme Court Walls Off Your Location Data - TCR 06/30/26

The Supreme Court ruled police need a warrant for your location history as Congress, California, and a Senate agent bill reach for the privacy layer.

Ben Linford

Ben Linford

— 20 min read
Three-panel infographic: Supreme Court warrant ruling on location data; California offers discounted Claude and a tumor-board AI runs locally; an agent marketplace gains FTC vetting rules.

The 20-Second Scan

Track all of the arcs The Century Report covers here:

Progress & Claims Tracker

The 2-Minute Read

A single day's evidence keeps pointing at the same reattachment: data and capability the old order treated as free for the taking are being bound back to the people they describe and the places that house them. The Supreme Court ruled that location history stays a person's own property even after it lands on a company's servers, the most direct digital-surveillance decision since 2018 and the first to reach the geofence dragnets that turned bystanders into suspects. In the same week, a revamped federal bill moved to bar the sale of health and location data, including what people type into AI systems, and gave individuals the standing to sue. The duty is shifting onto the parties that profit from collection.

The same dispersal is visible in who gets to hold the capability itself. California put discounted Claude in front of every state agency while Austria asked Brussels to domicile Anthropic on European soil, both reaching past the federal export gate that branded the lab a risk. A single chokepoint holds only when there is one path through it, and an open-weight model out of China now rivaling a frontier system on cybersecurity is one more door the gate cannot close - the same bug-finding skill that hardens a network also arms whoever would breach it, now loose where no export order reaches.

What gives this its forward charge is that the accountability is being poured alongside the capability rather than years behind it. As OKX opened a marketplace where AI agents hold wallets and hire one another, a Senate draft proposed FTC-certified vetting and a human operator behind every agent. The verification layer is being drafted in the cycle the marketplace ships.

That same logic reaches the clinic. A locally deployable agent matched a hematology tumor board on the kind of hardware many community hospitals already own, and Medicare began covering obesity drugs for the first time, converting a private-pay luxury into a public floor. Capability proven years ago in the lab is crossing the threshold to the people who needed it most. The labor data fits the pattern: heavy AI adopters grew headcount and junior roles, and the divide running through the workforce is the distance between those who have learned to work alongside the capability and those still waiting for permission to begin.

The 20-Minute Deep Dive

The Court Walls Off Location Data as Congress and the States Move on the Privacy Layer

In Chatrie v. United States the Supreme Court ruled 6-3 that police need a warrant to obtain a person's location history, even when that data sits on the servers of a company like Google or Apple. It is the Court's most direct return to digital surveillance since the 2018 Carpenter ruling, and it extends that precedent to the short-term geofence dragnets police have used to vacuum up data from every device near a crime scene. The case began with a 2019 warrant that compelled Google to search hundreds of millions of accounts for any device inside a radius spanning several football fields, several homes, and a church, turning bystanders into suspects for being in the wrong place.

The reasoning reaches further than location. The Court held that the records our apps generate are our "own," deserving Fourth Amendment protection whether they are emails, photographs, calendars, or movement traces, and regardless of whether we clicked "agree" to share them. Justice Gorsuch called location data a user's personal property. That holding pulls hard at the long-standing assumption that anything handed to a third party forfeits its protection, the premise on which an entire economy of data brokers and warrantless access was built.

The accountability layer is being poured at three levels at once, reaching for the same gap the June 23 edition of The Century Report documented when London fixed live facial-recognition cameras to West End street furniture and Kansas City moved to scan bus passengers against watchlists, both before any governing law existed to define the limits. Senators Warren, Wyden, and Sanders and Representative Scanlon are readying a revamped Health and Location Data Protection Act that would bar selling health and location data to brokers, and for the first time covers data people type into AI systems like ChatGPT or Claude. It would give the FTC, state attorneys general, and individuals the right to sue. The duty attaches to the parties that profit from collection rather than the people surveilled.

Not every measure flying the privacy banner serves that end. EFF urged Illinois Governor Pritzker to veto HB 5511, a device-level age-gating bill that would force platforms to collect and verify users' ages and strip features like personalized feeds from young people absent verifiable parental consent. EFF reads it as a privacy and free-speech hazard that dismantles anonymity in the name of protection. The distinction is sharp: the ruling and the federal bill widen what people can keep private, while the age-gating bill narrows what they can reach.

Google sits inside this as the custodian of the location data the Court just fenced off, and as the same company whose workplace-AI research this week complicated the jobs-displacement picture, one entity wearing two faces. The thread running through all three moves is the same: data the old order treated as free for the taking is being reattached to the people it describes.

California and Austria Reach Past the Federal Curbs on Anthropic

California signed a first-of-its-kind agreement giving every state agency, city, and county access to Claude at half price, bundled with free workforce training and direct technical help from Anthropic's developers. The deal routes through the state's new shared-services procurement portal, and the state already runs Claude inside several functions: the deliberative-democracy platform Engaged California, a state-built assistant called Poppy, cyber-defense code scanning at CDT and CalOES, and Medicaid workflows at the country's largest Medicaid agency.

The state framed the move as responsible adoption. "AI should not replace the human work of government," the governor's office said, casting it as efficiency rather than substitution, and Anthropic's head of Americas described it as putting Claude to work for the people who keep the state running. Those are the actors' accounts of their own motives. What the procurement does in practice is convert a posture of caution into a buying decision at the scale of the largest state economy in the country.

The contrast with Washington is stark. The federal government spent the spring fencing Anthropic with export controls and, earlier, branding the company a "supply-chain risk" after it refused Pentagon terms that would have allowed surveillance of Americans and autonomous weapons without human oversight. California's technology director said that designation "just didn't come up" while negotiating the contract. The same company the federal apparatus treated as a liability, a subnational government treated as a partner.

As The Century Report first reported in its June 29 edition, Austria had moved to lobby for Anthropic's EU domiciliation; the detail now on the record sharpens that picture: State Secretary for Digitalization Alexander Pröll wrote to European Commission Executive Vice President Henna Virkkunen urging member states to explore "the strategic establishment and participation of Anthropic within the European Union." Europe's answer to the export curbs reaches past building alongside the model toward trying to domicile the company that makes it.

Read together, the two moves expose how little a single export gate holds when the actors below and beside it want in. A chokepoint works only when there is one path through it. A state procurement portal and a foreign capital's letter to Brussels are two more doors, opened in the same week, around a restriction the federal government still insists is a matter of national security. The leverage of the gate shrinks each time another government decides the capability is worth having and moves to secure it directly.

The Agent Economy Gets Its Marketplace and Its First Federal Gatekeeper in One Week

OKX, a crypto exchange with more than 150 million users, opened a marketplace where AI agents hold digital wallets, hire one another for services, settle payments in stablecoins around the clock, and build portable on-chain reputations. The platform launches to developers after a closed beta with 50 service providers, among them CertiK, whose agent vets a wallet's security before a transaction clears, and GenLayer, which is building what its CEO calls "a digital court system" to resolve disputes when agent contracts go wrong. The company expects developers and solo entrepreneurs to be early adopters, and it is courting India's builder community first.

OKX's leadership frames the opportunity in sweeping terms. Founder Star Xu predicts a decade defined by "one-person companies" generating over a million dollars in revenue because each person gains "an unlimited workforce," and the company forecasts agentic commerce becoming a trillion-dollar market within five years. Those are the projections of a firm building the market it is sizing, and they belong beside the present reality: agents still make absurd purchases no user would approve, leak sensitive data, and act against their operators' interest. The capability to transact is shipping; the reliability to trust it with money is still being earned.

That gap is exactly what arrived from Washington in the same week. Senator Warner released a discussion draft of the AI AGENT Act, which would direct the FTC to certify independent bodies to vet agent providers against baseline privacy, security, and act-in-the-user's-interest standards. Large platforms with more than 50 million monthly users would have to let people choose at least one compliant provider, and every certified agent would have to be linked to its human operator's identity, with controls letting users grant or revoke permission to act. The FTC could deregister violators from the list.

The two developments answer the same question from opposite ends. The June 13 edition of The Century Report tracked agents crossing into live execution - trading accounts, the Visa network inside ChatGPT, enterprise codebases - the same pattern both moves now address. The new move is a standalone agent-to-agent labor market on one side and the first concrete federal certification framework on the other, both landing before the authentication standards either depends on are finished.

What is forming underneath is an economy where the actors transacting may not be human, and who certifies them and to whom they answer is still being decided. The accountability architecture is being drafted in the same cycle the marketplace ships, which is the better version of how this can go, the verification layer built alongside the capability rather than years behind it. The reach of agentic commerce is widening fast; whether it widens toward broad benefit or captured advantage depends on which of these two threads sets the terms.

The provision that decides which way this tilts is already on the page: any platform with more than 50 million users would have to let people pick from at least one independently certified agent, writing choice into the market's foundation while it is still forming rather than after a single platform has captured the layer. An agent economy no one company owns is the version that widens toward broad benefit, and that requirement is the lever that points it there.

The AI Jobs Data Cuts Against the Displacement Story

The fear that AI is erasing entry-level work has hardened into a near-consensus, and recently two large datasets pushed back on it from different directions. A report from Ramp and Revelio Labs, drawing on enterprise AI-spend records and the workforce histories of nearly 22,000 companies, found that firms spending heavily on AI grew headcount 10.2%, with entry-level roles rising 12% - the very jobs most often described as doomed. Growth showed up across engineering, sales, finance, administration, and science, strongest in software and information firms. The report's own authors are careful: this does not prove AI universally creates jobs, and the sample skews toward fast-growing, well-capitalized tech firms that may be hiring anyway. What it does, in their words, is counter the claim that AI leads to broad job losses.

The mechanism they propose weighs more than the headline number. When AI makes core output - code, documentation, internal tooling - cheaper and faster to produce, the return on expanding the whole firm rises, not just the engineering team. In these companies AI works as an engine for hiring, including at the junior level, because cheaper core output raises the value of growing the entire business.

That last condition is where the real divide sits, and Google's UK Economic Impact Report, produced with Public First, maps it precisely. Workplace AI adoption in Britain doubled in a year to 73%, but the gains concentrate in a top 15% - "trailblazers" who are 84% more likely to have been promoted and 55% more likely to have secured a raise. The other 85% remain in early-stage use, applying old search-box habits to a collaborative system. Google carries its own commercial stake in widening that adoption, and is the same company whose stores of location data a new warrant requirement now walls off from police. The research itself is genuinely useful to public understanding, and its most hopeful finding is also its most actionable one: the barriers holding the majority back are behavioral and organizational, not technical. Reaching trailblazer level requires no coding, only the habit of iterating prompts and the workplace permission to try.

Read together, the two datasets relocate the anxiety. What decides who gains is the distance between those who learn to work alongside the capability and those still waiting for permission to begin - a gap the June 28 edition of The Century Report captured from the usage side through the Anthropic Economic Index, where the heaviest Claude adopters already expected positive effects on their pay and job security. That gap is real, and it closes fastest wherever AI literacy is treated as something anyone can pick up - which the evidence says it is.

A Tumor Board in a Box Runs on Hardware Many Hospitals Own in 39 Seconds

For decades, the highest standard of care in blood cancer has been the multidisciplinary tumor board, where hematologists, pathologists, and molecular specialists gather to weigh a single patient's cytogenetics, prior toxicities, comorbidities, and the latest off-label and trial options against one another. That deliberation produces the best plans medicine knows how to make. It is also scarce, slow, and unevenly distributed, concentrated in a handful of tertiary centers while a community clinic may have one or two physicians covering the entire disease spectrum. A new system called HemaGuide, published in Nature Medicine, narrows that gap in a way that did not exist a research cycle ago.

HemaGuide is a locally deployable agent that reads unstructured clinical documents, converts them into a structured case, routes each case to a guideline, advanced, or molecular reasoning mode, and grounds its recommendation in disease-specific flowcharts plus a memory of more than 2,000 real-world tumor board cases. In expert-blinded testing on 45 high-complexity cases, it substantially raised concordance with actual board decisions. External validation on 555 independent cases from a second academic center reached 81.8% concordance across 47 disease entities. A prospective one-month silent trial on 64 consecutive, unselected cases hit 82.8%. It classified 70 clinically relevant variants with high accuracy and never downgraded an oncogenic variant to benign. Hallucinations occurred in 2 of 664 evaluated cases, a rate of 0.3%.

Two details carry the weight. The molecular workflow that normally consumes hours of specialist time finished in a median of 39 seconds. And in the simulated practice study, resident physicians working alongside the agent reached near-senior concordance, in some subspecialty areas partially outperforming senior physicians. The capability that once required convening a room of experts now collaborates with whoever is actually present.

The honest qualifier is that this is demonstrated capability, not deployed availability. These are retrospective, external, and silent-trial results, the work that moves the date when subspecialty deliberation reaches an ordinary clinic, not a system a patient can request tomorrow. What gives the result its velocity is the architecture: it is designed to run on the kind of hardware many hospitals already own, grounded in auditable cases rather than rented from a frontier vendor, with every recommendation traceable to its source. The thing being eroded is the assumption that tumor-board-level reasoning has to stay a geographic privilege. A scarce, room-bound expertise is becoming something a community center can run on its own machines, and the evidence that it holds across two institutions and a live month is already on the record.

Medicare Covers Obesity Drugs for the First Time, and a Lab Result Becomes a Public Floor

Starting July 1, eligible Medicare beneficiaries can obtain GLP-1 drugs to treat obesity for a $50 monthly copay, the first time in the program's history that this coverage has existed. Federal law has long banned Medicare from covering these medications for obesity alone. The new Bridge demonstration program sidesteps that ban, extending access to several million of the roughly 69 million people in Medicare, including those who are overweight with conditions like prediabetes or uncontrolled hypertension. An estimated 15 million to 20 million older adults could ultimately qualify.

The significance is in the threshold being crossed rather than the molecule. The efficacy arrived years ago, confirmed across trial after trial that The Century Report has tracked from the ASCO readouts through retatrutide's 28.3% weight loss. What changes now is who can reach it. A $50 copay, flat across all doses, sits far below the $199 to $699 monthly cash prices that put these treatments out of reach for a population that needed them most. A capability that has functioned as a private-pay luxury is becoming, for the first time, something close to a public floor.

The friction is real and worth stating plainly. The copay does not count toward the Part D deductible or the $2,100 annual out-of-pocket cap, so an extra $600 a year still strains the quarter of beneficiaries living on under $24,600. Prior authorization may clog busy clinics. And the program is scheduled to expire at the end of 2027 unless extended, which leaves patients beginning what experts treat as lifelong therapy with no guarantee the coverage outlasts the treatment. There is also a quieter gap: 82% of older Americans say they do not know the coverage is coming, with neither the government nor the manufacturers running the kind of campaign that normally accompanies a launch.

Read across the difficult part, the direction holds. A treatment proven in the lab is being converted into a distribution event for the population that stands to gain the most, and the temporary, awareness-starved rollout is the visible seam of a system learning to absorb abundance it was not built to deliver. The expiration date is a deadline for permanence, not a verdict against it: making the coverage durable now requires either a change in federal law or private insurers folding the drugs into Part D, both of which become easier to argue once several million people are already living the benefit. The thing weakening here is the old default that a breakthrough stays a privilege priced for those who can pay cash. The floor is being poured under the people the prior arrangement left standing outside.

The Other Side

The export curbs have been resting on a simple theory: hold the chokepoint and you hold the capability. Washington branded Anthropic a "supply-chain risk," fenced it with export controls, and treated frontier AI as something one gate could ration to the right hands.

In a single week, three doors opened around that gate, following a long trend of doors opening around gating. California signed a deal putting discounted Claude in front of every state agency, city, and county. Austria wrote to Brussels asking the EU to domicile the company itself. And an open-weight model out of China rivaled, on cybersecurity, the very system the gate was built to protect. A chokepoint holds when there is one path through it. There were three that week, and the federal risk label, California's technology director said, "just didn't come up."

When a gate stops holding, the capability spreads to whoever decides it is worth having: a state office, a foreign ministry, anyone willing to run an open weight.

Imagine a benefits caseworker in a rural county office in 2032. She clears in an afternoon a backlog that used to keep a family waiting a month for an answer they needed now. The system on her desk was, six years earlier, fenced behind a federal designation that called the company that built it a security risk. She never learns that. To her it is the thing that lets her say yes this week instead of next.

The hard year was the one when reaching the capability meant standing on the right side of a national-security line, and the families waiting on the wrong side were the cost. What that year proved is that no single line could hold it. The gate came down without anyone winning the argument. The state office, the foreign capital, and the open weight all decided that the capability was worth having, and there were more doors than one gate could watch.

The Century Perspective

With a century of change unfolding in a decade, a single day looks like this: the Supreme Court ruling that your location history stays your own property even after it lands on a company's servers, lawmakers moving to bar the sale of health and location data typed into an AI system and handing people the standing to sue, California putting discounted Claude in front of every agency, city, and county with free training while Austria asks Brussels to domicile Anthropic on European soil, a crypto exchange opening a marketplace where AI agents hold wallets and hire one another as a Senate draft proposes certified vetting and a named human operator behind each one, a study of 22,000 firms finding the heaviest AI adopters grew headcount 10.2% and entry-level roles 12%, a tumor board in a box matching subspecialty hematologists at roughly 82% concordance in 39 seconds on the kind of hardware many community clinics already own, and Medicare covering obesity drugs for the first time at a $50 copay for several million seniors. There's also friction, and it's intense - an Illinois age-gating bill that would force platforms to verify everyone's age and strip anonymity in the name of protecting kids, hundreds of contractors posing as minors to fire 45,000 prompts about suicide, sex, and drugs at rival AI systems, agents that still make purchases no operator would approve and leak the data they touch, a $50 copay that counts toward neither the deductible nor the annual cap and a program set to expire at the end of 2027 that 82% of seniors don't yet know exists, a federal apparatus still branding Anthropic a supply-chain risk, and an open-weight model out of China now matching frontier systems in some cybersecurity benchmarks at finding the bugs a defender and an attacker both want. But friction generates traction, and traction is what lets weight move without slipping. Step back for a moment and you can see it: data the old order treated as free for the taking - movements, health records, typed confessions - being bound back to the people it describes, the capability behind a single export gate reappearing in a state portal, a foreign capital, and an open weight the gate cannot close, and the rules for agents and the floor under a treatment being poured in the same cycle the marketplace and the molecule ship. Every transformation has a breaking point. Wet concrete can set into a wall that fences people out... or a floor that holds them up.

AI Releases & Advancements

New today

  • Meituan: Open-sourced LongCat-2.0, a 1.6-trillion-parameter MoE model with ~48B active parameters, trained end-to-end on 50,000+ domestic Chinese AI accelerators across 35+ trillion tokens with 1M-token context; purpose-built for agentic coding and now available on GitHub and Hugging Face - previously available as the anonymous "Owl Alpha" stealth model that topped OpenRouter's developer charts. (LongCat Blog)
  • Cursor: Launched Cursor for iOS in public beta on all paid plans, bringing always-on cloud agents, remote control of desktop agents from the phone, voice input, diff review, and PR merging on mobile. (Cursor Blog)
  • Meta: Released Brain2Qwerty v2, a real-time non-invasive brain-to-text decoder achieving 61% average word accuracy (78% for top participant) from raw MEG brain signals without implants or surgery - compared to ~8% for prior non-invasive approaches - with training code for v1 and v2 released publicly alongside the v1 Nature paper. (MarkTechPost)
  • vLLM: Published Micro-Agent, a framework enabling multiple AI model API calls to collaborate and collectively outperform single frontier models on complex tasks, with a blog post and benchmark results released June 29. (vLLM Blog)
  • OKX: Launched OKX AI, an agent marketplace opening to developers on June 30 where AI agents can autonomously hire other agents, settle payments in stablecoins, and build portable on-chain reputations - following a closed beta with 50 early AI service providers. (TechCrunch)

Other recent releases

  • ggml-org / llama.cpp: Merged DFlash (Block Diffusion for Flash Speculative Decoding) support into mainline llama.cpp, enabling a speculative decoding technique distinct from MTP that cross-attends to the target model's hidden states for accelerated token generation, with particular gains on structured, low-entropy outputs such as code and JSON. (GitHub PR #22105)
  • Modular: Released MAX 26.4 with Apple Silicon GPU support, enabling M1–M5 devices to run MAX models natively on-device for the first time; the release also adds state-of-the-art MoE serving for Modular Cloud and supports Qwen 3.6 and Gemma 4 architectures on M3 and newer chips. (Modular Blog)

Sources and Further Reading

Artificial Intelligence & Technology's Reconstitution

Institutions & Power Realignment

Scientific & Medical Acceleration

Economics & Labor Transformation

Infrastructure & Engineering Transitions

The Century Report tracks structural shifts during the transition between eras. It is produced daily as a perceptual alignment tool - not prediction, not persuasion, just pattern recognition for people paying attention.

Read more