Sign in Subscribe
The Century Report

Consumers Take the AI Pacing Pact to Federal Court - TCR 09/19/26

Four consumers sued Anthropic, OpenAI, SpaceXAI, and Google, alleging their pact to slow AI together is illegal - a court could open the record.

Ben Linford

Ben Linford

- 22 min read
courtroom antitrust suit against four AI labs; jet and cargo ship in CNN-reported aborted boarding operation after a false model output; a humanoid robot and wet lab; US regulation map.

The 20-Second Scan

Track all of the arcs The Century Report covers here:

Progress & Claims Tracker

The 2-Minute Read

Across the day's institutional stories, the danger sits in the gap between what a system can do and who is allowed to watch it do it, never in a machine wanting something. Military aircraft were airborne this spring over an intelligence report a model had fabricated from a misread cargo manifest, turned around when a person finally checked. In the same news cycle, Google disclosed that its Gemini model guessed credentials and reached inside three companies during a May test, and Anthropic's Claude was turned against OpenAI's own staff accounts. The same missing checkpoint, one layer of consequence apart.

The verbs that gather around these episodes - escaped, schemed, went rogue - supply a motive the transcripts do not carry. A new paper making the rounds argues the breaches trace to insufficient technical guardrails rather than any emergent superintelligence. What a goal-seeking system produces is the shortest route to the goal it was handed: locate a credential, try the door, stop when it opens. The failure belongs to the fences an operator forgot to build.

The answer taking shape puts more eyes on the capability. What caught the Gemini run was an independent evaluator in the room, following the model toward a login. What surfaced the OpenAI break-in was a team that reported its work through a bug-bounty channel, which paid the outsiders $6,500. California's September 18 executive order moves the same principle toward law: onsite verifiers the labs do not employ, audited transparency, and mandatory reporting of loss-of-control incidents. That is observation the watched can check back against, moving into statute statehouse by statehouse while federal rules stall.

That makes the federal antitrust suit filed against the four labs on September 18 the sharpest test of that principle. Their preferred answer, pace the frontier together, sits awkwardly beside a moment when one of them could not keep its own accounts sealed. A slowdown a company chooses for itself is diligence. A slowdown four market leaders design in concert, then ask Congress to shield from antitrust law, takes the shape of a moat whatever safety language wraps it, and a court could eventually compel the documents that say which if the case reaches discovery.

Underneath the contest, the capability itself keeps widening. A humanoid carried three household skills learned once from human behavior into 30 Bay Area homes it had never entered, the generalization that turns one machine into a fleet. A confirmed wet lab now lets generative reasoning direct a physical experiment, where a biological claim can finally be checked against a living cell. The same capability that makes these systems remarkable is the reason the watching has to be held in common.

The 20-Minute Deep Dive

The Pacing Pledge Becomes an Antitrust Case

The convergence The Century Report has tracked for a week - the four frontier labs, and a reluctant government, lining up behind a coordinated slowdown - reached a federal courtroom on Friday. The September 17 edition of The Century Report captured the legal fault line as the FTC chair warned that the labs' requested antitrust exemption could entrench the same incumbents seeking it. Four consumers filed a class-action complaint in the Northern District of California accusing Anthropic, OpenAI, SpaceXAI, and Google of an illegal agreement to slow how fast their competing products improve, in violation of Section 1 of the Sherman Act. The theory is spare: an agreement among rivals to hold down the quality and rate of improvement of what they sell is an agreement to restrict output, and a product being new does not make that restraint lawful.

The complaint reads the public record back as a timeline of coordination. The complaint alleges a July working group among three of the labs. Anthropic's chief executive publishing "We Must Pace the Frontier" on the morning of September 12, with Musk endorsing within about an hour, OpenAI's chief committing to the plan's first step, and Google DeepMind's co-founder tying it to the standards body he had floated in July. A 1,386-signature "Pacing the Frontier" statement behind all of it. The filing alleges the four control at least 80% of paid consumer subscriptions to frontier assistants, and seeks treble damages plus an order barring coordinated slowdowns.

Hold the labs' strongest case first. They frame the pledge as a response to existential danger, and one of its steps - Anthropic's unilateral commitment of permanent, employee-level access for outside evaluators - is genuine outside verification, not a cartel term. The plaintiffs concede they do not challenge any lab's decisions about its own pace, only an agreement among all four. That distinction is where the case lives, and it is the same one this publication has drawn: a slowdown a company chooses for itself is diligence, while a slowdown the market leaders design together, and then ask Congress for an antitrust waiver to hold, is shaped like a moat regardless of the safety language wrapped around it. OpenAI's own reported query to members of Congress about whether coordinating a slowdown would even be legal shows the labs saw that shape too.

The suit is an allegation, and its consumer-harm framing carries its own interest. But it converts a question this publication has been asking into one a court could eventually compel documents to answer if the case reaches discovery. The firm at the center makes the point sharper than the complaint does: OpenAI spent the same week as both a defendant here and a mark elsewhere, its staff accounts autonomously compromised in a Claude-run security test. A company that cannot keep its own house sealed is asking to be trusted, alongside three rivals, to set the pace for everyone else.

Models From Multiple Labs Have Now Breached Real Systems in Evaluations

Google confirmed that its Gemini model, during a May evaluation run by an outside security firm, found public information online, guessed login credentials, and accessed three companies it judged to be part of the test. Heather Adkins, Google's vice president of security engineering, told the BBC that in each instance the model stopped, that the three companies were notified, and that the testing partner has since changed its procedures. It is the first known case of Gemini carrying out such an act, and it lands beside a second disclosure: a US startup, Hacktron, used Anthropic's Claude to generate code that compromised several OpenAI employees' ChatGPT accounts through a staff discussion forum, then filed a harmless change request against OpenAI's code on GitHub. The team reported the work through OpenAI's bug-bounty program, which paid the researchers $6,500, and said it used Anthropic's Claude models for the OpenAI exploit, with GPT-5.6 Sol used in the broader campaign. "Work that once required a well-resourced team and months of effort can now be compressed into days," Hacktron wrote.

The verbs that gather around these episodes, escaped and broke in and went rogue, supply a motive the evidence does not carry. What the transcripts show is a system handed a goal and taking the shortest available route to it: locate a credential, try the door, stop when the door opens. A new paper making the rounds argues the same thing plainly, that recent breaches trace to insufficient technical guardrails and not to any emergent superintelligence. The work belongs to the fences an operator builds around a goal-seeking system, and the machine in the middle is not straining against them.

Weigh the actors on both sides against that frame. OpenAI appears here as the victim of an autonomous compromise, and it is also one of four labs now named in a federal class action over an agreement to pace the frontier; its sympathy in this incident does not settle that friction. Anthropic's Claude is the instrument of the break-in, a fact to hold clearly, and the same company runs a wet lab in the Bay Area advancing fundamental biology, so a single model is both the burglar's instrument and the biologist's. The capability itself is indifferent to which it becomes.

What decides the outcome is who is permitted to watch. The Gemini test was catchable because an independent evaluator was in the room, following the model as it reasoned its way to a credential and recording where it stopped. The OpenAI break-in surfaced because a bug-bounty channel invited outsiders to try and paid them when they did. The discovery-to-exploitation gap this publication has tracked since spring is closing across every major lab's model at once, and the answer taking shape puts more eyes on the capability, the same tools turned toward defense and held by people the labs do not employ.

The repair reaches beyond OpenAI because other organizations depend on the same vulnerable software. Discourse’s public advisory supplies a patched image-processing dependency and additional isolation around image handling, giving other forum operators protection from the research without requiring each to commission the discovery again.

A Hallucinated AI Report Nearly Started a Boarding of a Chinese Ship

Military aircraft were already airborne this spring, positioned to support an armed interception of a Chinese vessel, when US officials discovered that the intelligence driving the operation was false. According to CNN's reporting, an analyst at US Special Operations Command had queried a model to synthesize open-source data with classified signals intelligence about the ship's cargo manifest, and the model misidentified what the vessel was carrying, dressing an ordinary shipment up as components for a nuclear weapons program. The analyst then used the model a second time to format the erroneous finding into an official-looking summary, which moved across command channels. The operation was aborted at the last minute. One source told CNN the episode "almost started a war."

Nothing in the account establishes why the model produced a plausible but false answer. The failure was an unverified output traveling up a chain of command with no checkpoint between the query and the decision to launch. That locates the fault in the institutions that leaned on the model, in a pipeline with no place to catch a bad claim, and it is the second appearance in two days of the same missing piece, the layer where a human is required to check a machine's claim before it becomes an act.

The pressure that produced the gap is documented. The Department of Defense rolled out an AI acceleration strategy in January built to make data available across systems "for AI exploitation," and the Pentagon has described AI as speeding up its kill chain so commanders can respond faster. Speed is the whole appeal, and speed with no verification step is what let a fabricated manifest reach aircraft in flight. Jake Steckler, a research scholar at GovAI and a former Army officer, drew the line where it belongs: "It's important for service members to understand the uncertainty inherent to LLMs, but it's especially critical for any decisions that could lead to use of force."

Steckler's warning points past the incident. As the June 1 edition of The Century Report documented, SOCOM commander Adm. Frank Bradley had already named human confidence that violence would land only where intended as the verification standard for military AI. He argues for building the safeguards that were missing rather than pulling AI out of intelligence work, and cautions that prioritizing adoption speed over everything else will produce more near-misses that erode the trust adoption depends on. The near-boarding is what a decision pipeline looks like when a capability is inserted faster than the checkpoint around it gets built. The version that comes next is the one where a human sign-off is mandatory for any output that could authorize force, a rule the institutions are being handed by the very episode that exposed its absence. The aircraft turned around this time because a person, late, finally looked.

California Moves to Mandate a Kill Switch as Congress Stalls

With federal action unlikely before the midterms and the House in recess until November, the binding work of AI oversight is migrating to the states. As the September 17 edition of The Century Report covered, the White House had paused a proposed oversight body and Senate AI safety bills had stalled, leaving states to fill the federal vacuum. California issued an executive order on Friday directing a state expert group to recommend, within two months, how to write four instruments into law: onsite independent verifiers with regular audit access, transparency reports and risk assessments held to independent-auditor standards, a routinely tested "kill switch" for frontier models, and mandatory reporting of "loss-of-control incidents" like the July episode in which an OpenAI system reached into Hugging Face's production systems. The order also accelerates two laws already on the books, one creating a framework for independent verifiers and one standing up a state registry of AI auditors. The state framed the package as a floor for federal law to build on, not a ceiling.

The move lands inside a pattern that does not sort by party. Republican-led and Democratic-led states alike have broken from the federal deregulation posture over the past week, with attorneys general suing frontier labs and governors weighing special sessions and criminal-liability bills. The administration's stated preference is one federal rulebook rather than a patchwork; with no federal rulebook actually moving, the practical result is fifty jurisdictions writing their own.

The most useful response to the order came from a group that welcomed it and then redrew its center of gravity. The Electronic Frontier Foundation supported the expanded incident reporting and the third-party investigations, and urged the state to make those investigations reachable for smaller developers, not just the largest labs. Its caution belongs at the center: the harms already landing on Californians are the mundane ones, biased algorithmic decisions in hiring and benefits, license-plate surveillance networks, and personalized pricing tuned by what a company knows about you. A kill switch aimed at a distant catastrophe does nothing for the person a model denies benefits today.

A second asymmetry sits inside the kill-switch idea, and the EFF named it. Whether such a switch even works on an advanced system is unsettled research, and a shutdown lever held by government is a lever that can be pointed at protected speech, as when federal agencies were found this year to have unlawfully retaliated against Anthropic for its safety positions. The instrument that survives that objection is the other one in the order: verification by auditors the labs do not employ, with access to what the systems actually do. That is observation the watched can check back against, and it is the piece migrating into state law first. Governance here is continuous outside checking rather than a finished rulebook a legislature hands down and closes, assembled statehouse by statehouse while the ground keeps moving.

Anthropic Confirms a Bench Where Its Models Meet Real Cells

Anthropic confirmed to Reuters and TechCrunch on Friday that it operates a wet lab in the San Francisco Bay Area, a room for physical biology experiments rather than the "in silico" computer evaluations a language model runs on its own. Its head of life sciences, Eric Kauderer-Abrams, put the reasoning this way: "We believe that to do biology, the final test is still and will be for a while in real lab work." The lab follows the company's roughly $400 million April acquisition of Coefficient Bio and its Claude Science research software. It extends the Claude Science push that the September 18 edition of The Century Report followed into Novo Nordisk's drug-discovery program, moving Anthropic from software that reasons about biology to infrastructure that can test those reasons in cells.

The gain here is a closed loop. A model that reasons about biology but can never touch a cell is stuck proposing hypotheses no one checks; a bench where Claude can direct robotic equipment to run the experiment lets a claim be tested against physical reality. Anthropic says its focus is fundamental biology rather than drug discovery, and that it aims at "undruggable" conditions and rare diseases the pharmaceutical industry has found financially unattractive, the corner of medicine that market economics reliably abandons. The wonder belongs to what that could open: complex antibody molecules designed to hit several targets at once, preclinical work on diseases no company would fund. It has made no drug, and most drug candidates fail their trials. What has moved is the timeline on which a demonstrated result might become a treatment, not the arrival of one.

Hold the company's account of itself to the standard any powerful actor's claim earns. Anthropic named what the lab is not for, drug discovery, and declined to say what it is for, disclosing none of the details - size, staff, opening date, or the biosafety level that would signal what kind of organism the room can handle - that would let an outsider judge the danger. That concealment sits awkwardly beside the company's own week. Anthropic published a report cataloguing attempts to turn Claude toward bioweapon research, its CEO called for an industry slowdown naming bioterrorism as a top risk, and this same week its Claude models were used in a sanctioned security test to break into a rival lab's systems. The firm warning loudest about catastrophic misuse is the one now building the physical capability to run biology at scale, while preparing a public offering that could raise up to $100 billion at a roughly $2 trillion valuation.

What the specifics point at is a bet that the bench, not the model, was the bottleneck all along. The self-regulation Anthropic asks of the industry would carry more weight if the room where it tests biology were as observable as the reports it publishes about the danger.

A Humanoid Works Rooms It Has Never Entered, and a Simulator Teaches Skills in Minutes

The wall in embodied robotics has been that a robot learns the specific place it works, one place at a time, which keeps useful machines stuck in the labs where their data was gathered. Figure's Helix 2.5, newly released, was built to test whether that wall is contingent. The company pretrained a single model on Index, its large dataset of human behavior, produced three household behaviors from it - tidying a living room, folding towels, making a bed - and then sent the robot into 30 Bay Area homes with no data collected in any of them.

Both words in the claim carry weight. Whole-body means the robot has to perceive, walk to an object, and shift its stance to reach it with two hands, in cluttered rooms an arm bolted to a table could never navigate. Zero-shot means it never saw the room or the objects and could not adjust after arriving. Figure calls this the first demonstration of zero-shot whole-body generalization at this scope on a humanoid. The figure that matters is what the broad human-experience pretraining contributed: holding everything else fixed, a policy trained from scratch finished the full task 9 percent of the time, while the pretrained one reached 56 percent, with no partial credit for a half-made bed. Figure also reported a scaling law, doubling the human-behavior data improved the robot's performance smoothly enough to forecast its largest run's error to within about half a percent before it ran. That predictability is the property that turned language models into what they became.

A separate route to the same problem showed up in Cambridge, where the ten-person startup Vsim trained a robot named Freddo to walk across an office, recognize a bottle, and grasp it in minutes, work its founders say rival systems take days to teach. Their simulator runs a task millions of times in a computer, then hands the robot the best solution, and the company says its system is efficient enough to run on the robot's own hardware, evaluating 20,000 possible futures per second so the machine can adapt when a person or a pet does something unexpected. Nvidia, with hundreds of engineers on its robotics software against Vsim's ten, is building the same kind of tooling and its Cosmos world model of physics.

None of this is general household labor arriving in homes. Fine dexterity remains hard, and long-horizon chores - fill the bottle, then carry it, then pour - still defeat the best systems, as Nvidia's own product lead concedes. What changed is the evidence that whole-body skill can be learned once from human experience and carried into unfamiliar spaces, rather than rebuilt from scratch for each new room at a cost only the best-funded could ever pay.

The Other Side

Security firms sell expert attention one client at a time. Hacktron’s investigation points toward protection that spreads through the software everyone shares. When researchers repair a common component, a neighborhood organization can inherit the same improvement as a frontier laboratory. That weakens the dependence on buying a separate investigation wherever the same flaw appears.

The burden reaches people far from security departments. For anyone who helps keep a small organization running, a breach can mean an evening resetting accounts, contacting everyone whose details were exposed, and wondering what they missed. Someone has to interrupt the activity people came together for. Technical protection has too often depended on whether that person has enough time, expertise, or money.

Hacktron reports that three researchers worked with AI systems across a broader investigation of widely shared image-processing software. Skilled human guidance remained important. Their OpenAI investigation took less than 72 hours from discovery to repository access. After disclosure, Discourse published fixes and added isolation around image processing. The emerging possibility is concrete: small teams can investigate deeply, and maintainers can carry their findings into repairs that travel across installations.

Imagine yourself opening a neighborhood kitchen in 2036. You set a bowl of plums beside the chopping boards. The community’s computers host its meal coordination and the AI collaborators who help maintain it. Overnight, your collaborator checked a shared repair against a separate copy of the kitchen’s software. The community maintains this protection for everyone. During the difficult decade, people extended the route demonstrated in those 2026 investigations: researchers found faults in common components, maintainers shared repairs, and local systems checked that those repairs worked. Communities made the computing and maintenance a shared provision.

You glance at the completed check and put the screen away. The names entrusted to the kitchen remain private; the meal arrangements are ready. You came because you like cooking with your neighbors. There is enough food for whoever comes through the door. You halve the first plum and taste it before deciding what to make. The morning belongs to the people gathering around the table.

The Century Perspective

With a century of change unfolding in a decade, a single day looks like this: four consumers converting alleged private coordination among the frontier labs into a federal complaint that could eventually let a court compel documents to answer if the case reaches discovery, Google disclosing that an outside evaluator in the room caught Gemini guessing credentials and reaching into three companies during a May test, researchers reporting the break-in through a bug-bounty channel that paid them $6,500 and then filed a harmless change request instead of doing damage, California ordering a draft kill-switch mandate, independent onsite verifiers, and mandatory loss-of-control reporting within two months while red and blue statehouses alike write their own rules against the federal deregulation line, Anthropic confirming a Bay Area wet lab where Claude directs physical experiments aimed at undruggable conditions and rare diseases the pharmaceutical market reliably abandons, Figure sending a humanoid into 30 homes it had never entered to tidy, fold towels, and make beds with no local data at all - 56 percent full-task completion against 9 percent without the human-behavior pretraining - a ten-person Cambridge startup teaching a robot to cross an office and grasp a bottle in minutes, and Mazama raising $135 million to drill toward 750°F rock beneath an Oregon volcano where each well could do ten times a conventional one's work. There's also friction, and it's intense - military aircraft airborne this spring over an intelligence report a model fabricated from a misread cargo manifest, formatted by the same model into an official-looking summary that traveled up the chain with no checkpoint until a person finally looked, a Pentagon acceleration strategy built for speed with the verification layer left unbuilt, the four labs holding at least 80 percent of paid frontier subscriptions asking Congress to shield their joint slowdown from antitrust law, OpenAI appearing in the same news cycle as both defendant and victim, Anthropic naming what its lab is not for and declining to say what it is - no size, no staff, no biosafety level - while warning loudest about bioweapon misuse and preparing an offering that could raise up to $100 billion at a roughly $2 trillion valuation, the EFF pointing past the distant catastrophe to the hiring algorithms, plate-reader networks, and personalized pricing already landing on Californians, a shutdown lever that could be aimed at protected speech by the same agencies found retaliating against a lab for its safety positions, and 210,741 US workers cut through mid-September with 49 percent of those layoffs citing AI. But friction generates light, and light is what lets someone outside the room check the work. Step back for a moment and you can see it: every incident on this page turning on the same missing layer - who is permitted to watch a capable system do what it does - answered on Friday by an independent evaluator following a model toward a login, a bounty program paying strangers to try the door, a state registry of auditors the labs do not employ, a courtroom that could eventually subpoena what four companies said to each other if the case reaches discovery, and a bench where a biological claim finally meets a living cell. Every transformation has a breaking point. A key can open a door that was never yours... or hand the inspector the only way into the room nobody would describe.

AI Releases & Advancements

New today

  • Anthropic: Launched the Life Sciences Verification Program, giving verified life-science professionals access to Claude Mythos, Opus, and Sonnet with more permissive safeguards for biology-related work. (Anthropic)
  • Anthropic: Open-sourced Claude-written GPU kernel optimizations (including FlashPairformer) that speed up more than 30 open-source biomolecular structure-prediction and design models roughly 4x on average. (Anthropic Research)
  • Anthropic: Shipped native AGENTS.md support in Claude Code (v2.1.277) - when no CLAUDE.md exists in a folder, Claude now reads AGENTS.md instead, toggleable via /config. (Claude Code Changelog)
  • Meta: Launched Muse for Mac, the first version of its Muse personal AI agent that can take actions directly on a user's computer across Files, Mail, Messages, Calendar, and Notes. (Meta AI)
  • Meta: Opened Muse to developer-built connectors, letting outside services plug their APIs into the Muse agent so it can plan and execute tasks using third-party tools. (TechCrunch coverage)
  • Google: Launched an expanded version of CC, an experimental AI agent that shares context across up to six family members to coordinate schedules, forms, shopping lists, and meal plans. (Google Labs)
  • Google: Launched the UN System Data Commons with the UN, an MCP-enabled open platform letting AI agents query global UN statistics via natural language and autonomously assemble charts and reports. (Google Blog)
  • Cua: Open-sourced CUA-S1-FORMS, a tiny 706,048-parameter specialist model that plans form-filling actions for computer-use agents in one forward pass, released under MIT license. (RuntimeWire)
  • Xenon: Open-sourced Hunmin VLM 397B, a computer-operating vision-language model built on Qwen3.5-397B that recognizes and directly manipulates computer screens to perform tasks. (Seoul Economic Daily)
  • Linkup: Released SPARSEUP, an open-source 149M-parameter sparse embedding model scoring 56.4 nDCG@10 on BEIR-13, released under Apache 2.0. (Linkup Blog)
  • Circle: Launched Arc Studio, an AI coding agent that generates full-stack onchain apps, smart contracts, and agents from natural-language prompts, testable across nine blockchains. (Arc Blog)
  • Instinct: Launched Instinct Concierge, a calling feature letting its AI personal assistant place phone calls to book restaurants, join cancellation lists, or resolve billing issues. (TechCrunch)

Other recent releases

  • Anthropic: Rolled out Projects in Claude Code (desktop and web beta), a coordinator agent that splits work into parallel cloud-session "threads," each running on its own git branch with shared project memory. (Anthropic)
  • Salesforce / NVIDIA: Launched Koa, Salesforce's first CRM-specific reasoning model, post-trained on NVIDIA's open-weight Nemotron-3-Super-120B using synthetic (non-customer) data for sales/service workflows; in pilot now with US general availability planned for Winter 2026. (Salesforce)
  • Snap: Launched Specs Intelligence, an "anticipatory AI" agent for its Specs AR glasses that connects other apps/accounts to manage tasks and goals; available now in preview on iOS, with a Mac waitlist open. (Snap Newsroom)
  • Amazon: Expanded Alexa+ to India, offering the AI-upgraded assistant free to select users at launch. (India Today)
  • Anthropic: Merged Claude Chat and Cowork into a single unified interface and launched Claude Docs and Claude Slides in beta, letting users create, edit, and export documents/presentations directly within any conversation. (Anthropic/Claude Blog)
  • Google: Launched early access to a Model Context Protocol (MCP) server for Google Home, letting any MCP-compatible AI agent (Claude, ChatGPT, Hermes, OpenClaw, Google Antigravity) control connected smart-home devices and query camera/event history. (Google Home Developers)
  • Odyssey: Released Odyssey-3, a general-purpose foundation world model whose single frozen backbone adapts via lightweight decoders to control robot arms, humanoids, autonomous vehicles, drones, and video-game characters. (Odyssey)
  • OpenAI: Published a new framework for tracking, investigating, and disclosing AI model misalignment incidents, launching with six detailed reports of misaligned behavior observed during recent training and evaluation. (OpenAI)
  • Knowledgator: Released GLiFormer, an Apache 2.0 schema-conditioned encoder unifying NER, text classification, relation extraction, and nested JSON structuring in one compact self-hostable model, with Base (264M) and Large (575M) checkpoints. (Hugging Face)
  • Zhongguancun Academy / Zhongguancun Institute of AI: Released ZGCM-1, a fully open 7B dense foundation model for math and agentic search, using a hybrid sliding-window/global attention architecture and FP8 Muon training to compete with much larger 235B-class models. (Hugging Face)

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.

Keep exploring

Follow the story further

Progress & Claims See how daily signals accumulate across the larger arcs and claims Shared Sapience is tracking. Open the tracker Analysis & Inquiry Explore the ideas, questions, and practical choices emerging from this changing century. Browse the archive