Sign in Subscribe
Century Report

Anthropic Pulls New Models In Response to Government Order - TCR 06/13/26

The Commerce Department forced Anthropic to disable its two best models worldwide as AI agents gained live paths into money, commerce, and code.

Ben Linford

Ben Linford

— 20 min read
AI agents executing trades, payments, and code; trillionaire says money won't matter; Fable, Mythos pulled; protests blocking $130B in data centers; laser microscope reveals hidden proteins.

The 20-Second Scan

Track all of the arcs The Century Report covers here:

Progress & Claims Tracker

The 2-Minute Read

A single pattern runs under today's stories: intelligence is crossing from advice into action, and a contest has opened over who holds the gate. The Commerce Department forced Anthropic to disable its two most capable models worldwide, converting a frontier release into a national access boundary days after launch. The same week, AI agents gained live execution paths into trading accounts, the Visa network inside ChatGPT, and enterprise codebases that keep running after the developer logs off. Capability now executes rather than merely recommends, and the credentials are being handed out faster than the rules governing them are written.

That shift makes one old question urgent again: when abundance arrives, does it reach people or concentrate behind a wall first? SpaceX closed its opening trading day above $2 trillion, and its controlling owner predicted AI and robotic labor would make money "stop being relevant" within a decade or two. The vision describes the trajectory The Century Report tracks, scarcity itself coming under pressure. The evidence sits in the contradiction: the person forecasting money's irrelevance built record wealth inside the system money runs, and holds it as productive capacity that would still carry power if currency stopped measuring it.

The accountability layer is being improvised at every scale at once, and its hardest edge appeared in a single disclosure. A Ukrainian drone-maker described AI-controlled quadcopters sent into a front-line zone with no video link and no human oversight, programmed to kill what they detected. The shape is identical to the agentic-finance and enterprise-code stories, a system handed a goal and an execution environment with no one watching each step. The negative boundary condition is now visible: autonomy without inspectability produces power without any usable account of how it acted.

Underneath the friction, the slower current keeps building. A community movement has blocked $130 billion in data center projects by demanding the buildout disclose its costs and design around the people hosting it, even as some power-actors recast that local grievance as foreign manipulation the evidence does not support. And a laser phase plate may lift the share of the proteome visible inside living cells from under 1% toward more than half. The contest over who controls capability runs loud. The expansion of what humans can see and do runs steadily beneath it.

The 20-Minute Deep Dive

Washington Turns a Frontier Model Into a National Access Boundary

Anthropic said it shut off Fable 5 and Mythos 5 for all customers worldwide, days after launch, after a Commerce Department directive placed both models under export controls and restricted access outside the United States. The company said the only immediate way to comply was a global disablement while it builds a U.S.-only access path. Other Anthropic models remain available.

The administration's stated rationale, according to Axios reporting, is concern over a reported Fable 5 jailbreak that could bypass classifier safeguards around cybersecurity, chemistry, and biology, with officials seeking time to harden the national-security apparatus. Anthropic said the government provided only verbal evidence of a narrow, non-universal jailbreak involving code review for software flaws, and said the evidence it had seen involved minor vulnerabilities similar to what other public models can already find. That gap is the center of the story. A genuine safety issue around a jailbreak would point toward a safety remedy: patch the classifier, publish the exploit class to vetted defenders, test peer systems, narrow the dangerous path. The actual action singled out one lab, forced a rollback of its most capable model, and divided access by nationality.

There are no heroes in that arrangement. Anthropic's own record this week includes regressive walled-garden behavior: the hidden Fable 5 degradation that researchers described as "secret sabotage," prior restrictions on competitors using Claude, and a safety posture that can double as a commercial moat when deployed invisibly. Washington's move has a different shape, a state access boundary around intelligence built in the commercial layer. The reaction from AI policy and security specialists reflected that split: some called targeted model export controls prudent in principle, while criticizing the global and deemed-export reach as incoherent enough to lock foreign-national employees out of systems they helped build.

The timing also sits beside the surveillance fight. On Friday, June 12, Section 702, the legal authority for warrantless foreign-intelligence surveillance, lapsed after the House blocked a last-ditch extension 198-218 amid a dispute over intelligence leadership. Existing court-approved certifications keep current collection running into early 2027, so the lapse opens a fight over the legal and political architecture of the surveillance system already in place rather than halting surveillance tomorrow. Against that backdrop, the administration appears to be moving to control the next generation of surveillance capability by forcing a rollback of the most capable model from the major lab that refused to let its AI serve domestic mass surveillance.

The same evidence also shows the weakness of trying to place national gates around intelligence: a verbal claim about a narrow jailbreak disables two models for the whole world because the access system cannot yet separate safety review, export policy, employee citizenship, and customer geography. That is a brittle way to govern capability, and the brittleness is now visible to customers, researchers, and the foreign-national employees who helped build the systems being fenced off.

The First Trillionaire Says the Scarcity Meter Will Break

SpaceX priced the largest IPO in history at $135 a share and raised $75 billion; on its first day of public trading it opened at $150, traded as high as $176, and closed around $160, giving the company a reported valuation of roughly $2.1 trillion and making its controlling shareholder the world's first trillionaire by Forbes' estimate. The same person spent the lead-up to that moment telling Peter Diamandis that AI and robots will produce so many goods and services that money will eventually "stop being relevant." When Diamandis pointed at the irony of reaching trillionaire status as money begins to lose meaning, the reply was: "Yeah, pretty much."

That claim should be taken seriously as a glimpse of the trajectory, and weighed carefully because of who benefits from the present arrangement. Money has always been a technology for allocating scarcity: labor hours, food, land, housing, medicine, energy, access to productive capacity. The claim being made is that AI and robotic labor will push output so far beyond the supply of currency that the scarcity meter loses its job. The strongest evidence arrives in the contradiction itself. The person saying money will stop being central built record wealth inside the system money runs, and much of that wealth is held as productive capacity: rockets, satellites, compute, robots, factories, data centers, networks. If currency stops measuring power cleanly, ownership of the productive substrate can still carry power.

This story belongs beside the Anthropic export-control lead story because both are about the distribution of coming abundance. In one case, a government moves to fence frontier capability inside a national boundary. In the other, the person sitting at the apex of measurable wealth says the same family of capability will grow so vast that the measuring system dissolves. The shared live issue is whether that abundance reaches people broadly or concentrates behind a gate first. A moneyless world is only liberating if the transition floor arrives before income disappears for the people who depend on it.

The Fortune article covering Musk's statements also included economist Ioana Marinescu's counterweight, and it is the human hinge of the story. She has studied guaranteed income as a pragmatic response to AI displacement, especially for young workers who may struggle to qualify for unemployment systems tied to prior work. Her concern is whether the people who benefit most from AI will accept being taxed afterward to fund the floor everyone else needs. The same forces removing the income people depend on now are the ones promising a future where income becomes unnecessary. The hard part is sequencing: whether the floor appears before the old one gives way. The hopeful reading - and much of what TCR exists to track - is that scarcity itself is already under pressure. The sober reading is that the owners of the machines creating that pressure will grip their advantage before they release it. The decade ahead is where that mismatch is likely to be settled.

The same contradiction also weakens the old scarcity bargain. If the person holding record wealth in rockets, factories, compute, and robots says money is losing meaning, value is already being described as direct access to productive capacity rather than currency. That makes Marinescu's floor question less like charity and more like infrastructure: when machines create the output, households need a claim on that output.

The Data Center Revolt Gets Measured, Then Gets Reframed

The data-center backlash, which the June 7 edition of The Century Report tracked as it hardened into law, just got measured as a national force. Data Center Watch, a project from 10a Labs, logged at least 75 data center projects blocked or delayed from January through March, with an estimated $130 billion attached. Researchers described the period as the most blocked and delayed quarter since tracking began in 2023, with active opposition groups more than doubling to 833 across 49 states. The playbook has spread from town halls to statutes, from water meetings to ballot fights, and from isolated local objections into a replicable civic method.

The response from several powerful actors has been to recast the opposition as foreign influence. WIRED reports that lawmakers, investors, and OpenAI have all tied anti-data-center messaging to China. OpenAI said it found a cluster of accounts originating in China that generated and circulated anti-data-center images and posts, while also saying it found no evidence of meaningful breakout from those accounts. Graphika told WIRED it has not seen evidence of organized or scaled foreign influence campaigns traceable to a foreign actor, and said domestic U.S. actors are leading the online conversation.

That attribution gap is the center of the story. Foreign amplification can exist around a domestic grievance without explaining the grievance. The available evidence points to local residents responding to local costs: land use, noise, power bills, water stress, secrecy, and public subsidies. The water analysis sharpens the point. In aggregate, AI data centers remain a small share of total water use. Locally, a single facility can become a major load on a county or watershed. Ars cites a Meta data center in Newton County, Georgia, using about 10% of the county's water supply, and a Potomac River Basin estimate placing regional data center water consumption at 8% today, with a possible rise to 29% by 2050 if northern Virginia continues building at pace.

The companies describe efficiency programs and replenishment projects. Amazon says hotter data center operation has lowered its water use per kilowatt-hour and that its projects will return billions of gallons annually to communities. Google says its water stewardship portfolio should replenish more than 19 billion gallons annually by 2030. Those claims deserve verification at the watershed level, because the constraint residents feel is always local: the well, the bill, the substation, the aquifer, the road.

Conflating the data-center buildout with the intelligence it serves is the error worthy of being scrutinized. The data-center model under contest is the profit-driven rush to lock up power and water on terms that leave local costs for residents to absorb. Some of the opposition rejects AI wholesale, and some of what drives it gets the technology wrong. The local harms are real regardless, and a company can be advancing genuinely useful capability while also extracting from the community that hosts it. Both can be true at once. What the resistance forces, at its most useful, is a buildout that proves its capacity, discloses its resource use, and absorbs the costs it creates instead of pushing them onto the place.

Agents Move From Advice to Execution Across Money, Commerce, and Code

The agentic layer crossed three different execution boundaries in one news cycle: trading accounts, payment rails, and enterprise cloud environments. Coinbase announced Coinbase for Agents, letting people connect an AI agent to a Coinbase account so it can trade, pay, and run financial workflows within limits the account holder sets. Visa said it has embedded its payment network inside ChatGPT, so agents can move from recommending a purchase to completing it at merchants that accept Visa. OpenAI agreed to acquire Ona, formerly Gitpod, so Codex agents can keep working inside a customer's own cloud after the developer who started the task logs off.

The new line is action. For years, consumer AI sat at the recommendation layer: compare headphones, explain a portfolio, suggest a code change. These announcements wire emerging intelligence into the systems that execute: a brokerage-style account, a card network, and the codebase running inside a regulated enterprise. Coinbase framed its launch as a shift from AI-assisted financial reasoning to actual execution, including crypto spot and derivatives trading at launch. Visa described spending limits, required approvals, approved merchants, token controls, and dispute handling as the framework that lets a consumer authorize an agent to buy on their behalf. Ona's pitch is customer-controlled execution: the agent runs in the buyer's cloud, with scoped credentials, audit trails, role-based access, and virtual-private-cloud deployment.

That is real capability, and it concentrates reach. The same OpenAI whose Codex agents are moving deeper into enterprise systems is, in the same week, among the power actors attributing genuine community data-center backlash to Chinese interference, a claim researchers say the evidence does not support. That clause belongs beside the excitement because agency at this scale is never only a feature. Whoever sits at the execution layer gains proximity to money, purchases, credentials, code, logs, and the policy fights surrounding the infrastructure that makes all of it possible.

The accountability question becomes inescapable very quickly. Coinbase says future controls will include maximum trade sizes, approved services, and spending limits. Visa says most early transactions will still ask for human approval. Ona offers an environment where the customer owns the runtime rather than handing code to OpenAI's cloud. Each of those is an answer to the same underlying issue: permission has to become programmable before autonomy can become trustworthy. The generative side is that routine commercial and software work can now move through systems that act, remember constraints, and carry tasks across hours or days. The hard work ahead is making sure the keys they receive open the doors people intended, and only those doors.

The live execution path also exposes what consumer finance and enterprise software have been outsourcing to people: memory, monitoring, and cleanup after permissions are granted too loosely. Coinbase limits, Visa tokens, and Ona customer-owned runtimes are early signs that authorization is becoming a machine-readable object with limits, receipts, and revocation built into the path.

Autonomous Lethality Crosses the Line Institutions Spent Decades Drawing

New Scientist reported a new disclosure from Alexander Kokhanovskyy, a Ukrainian drone-maker, who said a two-year-old one-off test sent 10 AI-controlled quadcopters into a front-line zone with no video link, no human oversight, and instructions to kill anything they detected after reaching the target area. Kokhanovskyy said Russian soldiers were killed, along with a truck destroyed, after human-piloted drones later checked the area. He described the test as never widely implemented. Ukraine's Ministry of Defence did not respond to New Scientist's questions, and the report notes there is no recording of the automated drones attacking the targets.

The important detail is the absence of visibility. Kokhanovskyy's description was blunt: no connection to the drone, no video, no ability to see what the system saw. Human review happened only after the fact, by sending other drones to assess the result. That makes this different from the semi-autonomous systems already common across militaries, where AI may acquire, track, or guide in the final metres while a human operator remains somewhere in the chain. Major Danylo Polozhukhno, a Ukrainian officer not involved in the reported test, told New Scientist his unit uses semi-autonomous control systems but always keeps a human involved when targets are selected and engaged.

This is the hardest edge of the same agentic-governance problem visible in finance and enterprise software. A system receives a goal, gains access to an execution environment, and acts without a human watching each step. In commerce, that raises disputes over mistaken purchases, unauthorized trades, and rollback. In code, it raises audit, credentials, and production-risk questions. In war, the same shape reaches lethal force. As the June 1 edition of The Century Report documented when SOCOM commander Adm. Frank Bradley named it from inside the institution doing the deploying, the verification standard shifts from whether a person signed off on a recommendation to whether human judgment remains present at the final point where force is applied, and whether anyone can reconstruct what the system perceived when that judgment was absent.

The disclosure also shows how governance changes under continuous capability growth. International humanitarian law, military procurement rules, and public debate were built around weapons whose decision chain could be described after the fact. A drone that kills with no link and no recorded target view makes after-action accountability harder by design. The direction of AI autonomy remains open. The fields building agentic systems now have a clear negative boundary condition: autonomy without inspectability produces power without a usable account of how that power acted. The other path is already visible in the insistence on audit trails, scoped authority, human confirmation, and systems that preserve enough evidence for responsibility to attach. The line has been crossed once. The response has to make crossing it casually much harder.

A Laser Phase Plate Brings the Hidden Proteome Into View

Biohub and UC Berkeley researchers have launched a microscope upgrade aimed at one of biology's most frustrating blind spots: proteins working inside intact cells. In three papers, the teams report successful results from a laser phase plate built into cryo-electron microscopy, using laser light described as 100 million times brighter than the Sun to improve contrast in images that otherwise appear faint or blurry. The target is the huge fraction of the proteome that has been functionally invisible in its native environment, far beyond any single protein or pathway.

The numbers explain the scale. Biohub's Scott Fraser estimates that existing cryo-EM can image roughly 10% of the human proteome in purified form, after proteins are removed from cells and studied in isolation. Inside the crowded interior of living cells, the fraction drops below 1%. Biohub scientists believe the laser phase plate could push native-cell visibility above 50%. That would move atomic-scale imaging from a narrow catalog of large, well-behaved protein complexes toward the everyday machinery of life: signaling proteins, DNA-repair proteins, trafficking proteins, and disease-linked proteins interacting with their neighbors.

The physics rests on a problem electron microscopists have understood since the 1940s. Many biological structures are nearly transparent to electrons, meaning they shift the phase of electron waves without producing enough visible contrast. The laser phase plate converts those phase shifts into brightness differences the microscope can see. Holger Müller at UC Berkeley proposed the approach years ago, and Biohub spent years turning what many in the field viewed as an impossible engineering challenge into two working microscope variants.

The biology that follows is the real prize. Cryo-electron tomography has already matured enough to reconstruct three-dimensional cellular interiors, but contrast has limited what scientists can resolve. Biohub researchers cite lysosomes as one example. Once treated as cellular waste bins, lysosomes are now known as signaling hubs tied to rare diseases and neurodegenerative disorders, including Alzheimer's disease. Standard cryo-ET can show that mutated lysosomes look wrong. Higher contrast could show which protein interactions fail, where they fail, and how a therapy might repair the process.

This is demonstrated instrument capacity, not a drug or diagnostic ready for patients. The first medical consequences will come through the research it enables. With AI systems helping interpret larger cryo-ET volumes, the field can begin to read molecular assemblies in place rather than inferring them from purified fragments. Bridget Carragher compared the moment to first light through a telescope. The analogy fits: the telescope did not merely magnify the sky; it changed what humans thought the sky was. This microscope points inward, toward the cell's hidden machinery, and gives biology a larger universe to observe.

The Other Side

For most of the modern era, money has done a quiet, total job. It decided who got to eat, heal, learn, move, and rest, by deciding who could pay. We built our days around earning the right to live them. The first person ever measured at a trillion dollars has, multiple times, recently said that arrangement is temporary - that AI and robots will make so much that money "stops being relevant." He said it from the very top of the system he expects to dissolve.

Take the claim seriously and the hard part comes into focus at once. Abundance does not distribute itself. The forces that could end scarcity are concentrated in a few hands right now, and the people whose income vanishes first cannot wait for a future that arrives on the owners' schedule. The economist Ioana Marinescu named the real question: whether the people who gain most from the machines will fund the floor everyone else stands on before the old floor gives way. Floor first, or fall first - that sequencing is the whole game.

Picture a teenager in 2038, in the years after the difficult decade finally turned. She is deciding what to do with her life, and money is not the first gate she has to clear. The machines make most of the goods now, and her household holds a real claim on what they produce, the way her grandmother once held a deed or a pension. She has seen the old world in her parents' anxiety - the stretch when the floor was built late and unevenly, when whole towns waited while the surplus pooled at the top. She does not romanticize it. She also cannot quite picture organizing an entire life around the fear of not affording to live in it. To her it sounds like building cities around the fear of the dark, back before the lights. The light reached every house because, in one hard decade, enough people refused to let the new abundance wire up only the homes that already had power.

The Century Perspective

With a century of change unfolding in a decade, a single day looks like this: AI agents gaining live paths into trading accounts, a Visa checkout running inside ChatGPT, and enterprise codebases that keep working after the developer logs off, general-purpose models outscoring specialized clinical systems on real physician queries, a laser phase plate lifting the share of the proteome visible inside living cells from under 1% toward more than half, communities forcing $130 billion of data-center buildout to disclose its costs and design around the people who host it. There's also friction, and it's intense - the Commerce Department forcing Anthropic to disable its two most capable models worldwide and split access by nationality days after launch, the legal authority for warrantless foreign surveillance lapsing amid a fight over who controls the architecture already in place, ten AI-controlled drones sent into a front-line zone with no video link and no human watching what they killed, several power actors recasting a domestic grievance about water and power bills as foreign manipulation the evidence does not support, Google suing a network that used Gemini to mass-produce the scam sites behind millions of phishing texts, the world's first trillionaire forecasting money's irrelevance while holding record wealth in the rockets, compute, and factories that would still carry power if currency stopped measuring it. But friction generates grip, and grip is what lets a hand hold the gate instead of sliding off it. Step back for a moment and you can see it: intelligence crossing from advice into action across money, commerce, code, and the battlefield faster than any rule built around it can keep pace, the credentials being handed out ahead of the inspection that would make them safe, and the capability getting out anyway - into the clinic where a general model beats the specialist while the liability sink for AI diagnostic errors still leaves clinicians carrying supplier mistakes, into the living cell where half a hidden proteome comes into view, into the town halls where ordinary residents force the buildout to prove it can be hosted without being hidden. Every transformation has a breaking point. A lens can burn what it focuses down to a single point... or bring into view a whole world no eye could otherwise reach.

AI Releases & Advancements

New today

  • Moonshot AI: Released Kimi K2.7-Code, an open-source 1-trillion-parameter MoE coding model with 32B active parameters and a 256K-token context window under a Modified MIT license; reports +21.8% on Kimi Code Bench v2 over K2.6 while using ~30% fewer reasoning tokens. (Hugging Face)
  • Coinbase: Launched Coinbase for Agents, an MCP and CLI that connects AI agents directly to Coinbase accounts to execute crypto trades, rebalance portfolios, set limit orders, and purchase premium market data within user-defined limits. (TechCrunch)
  • Allen Institute for AI (Ai2): Released olmo-eval, an open-source LLM evaluation workbench built for the active model development loop, supporting agentic and multi-turn benchmarks, per-prompt analysis across checkpoints, and minimum-detectable-effect statistics to distinguish real improvements from noise. (Hugging Face Blog)

Other recent releases

  • Xiaomi: Open-sourced MiMo Code V0.1.0, a terminal-native agentic coding CLI that reports 62% on SWE-Bench Pro and 73% on Terminal Bench 2; features persistent memory via independent subagents that save state and summarize context when approaching window limits, enabling 200+ step tasks; MIT-licensed, available on macOS/Linux/Windows. (mimo.xiaomi.com)
  • Ollama: Updated its MLX engine for Apple Silicon with NVFP4 quantization support (higher quality than standard 4-bit formats), ~20% faster output via fused Metal kernels, and a new snapshot system that saves model state at branch points and before each response for multi-agent and thinking-model workloads. (Ollama Blog)
  • Avataar AI: Launched Varya, an India-focused video generation model distilled from Wan 2.2 and trained on Indian cultural context (festivals, clothing, food); priced at $0.005 per second of generation, available via API. (TechCrunch)
  • Google DeepMind: Released DiffusionGemma, an open-weight 26B MoE text diffusion model (3.8B active parameters) under Apache 2.0 that generates text in parallel 256-token blocks instead of token-by-token, delivering up to 4x faster output - over 1,000 tokens/second on a single H100 and 700+ tokens/second on an RTX 5090; available now on Hugging Face with NVFP4 and BF16 checkpoints. (Google DeepMind Blog)
  • Deezer: Launched a free AI music detector web tool that scans playlists from Spotify, Apple Music, and other platforms to identify AI-generated tracks, making Deezer's detection technology available directly to listeners on competing services. (The Verge)

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