The Century Report

China Rejects as "Fearmongering" Calls for an AI Slowdown as Open Models Spread - TCR 09/15/26

Beijing dismissed the AI slowdown as a 'Cold War playbook' while its spy chief warned of threats to Party rule, even as open models keep spreading.

Ben Linford

15 Sep 2026 - 22 min read

The 20-Second Scan

Beijing's foreign ministry called Amodei's AI-slowdown essay 'fearmongering' after it urged Washington to impede China, state media branded it a 'Cold War playbook', and the state-security minister named AI a sixfold threat to Party rule.
A Louisiana senator readied a bill requiring every AI model to carry a kill switch and a House Democrat pressed for a global AI framework as a cross-party bloc broke with the president's 'HOAX' dismissal.
Europe must build its own AI models and datacentres or risk being cut off by the US or China, the ECB president said, as Germany called an AI slowdown 'not viable' for digital sovereignty.
Definium's single-dose LSD therapy DT120 cut anxiety scores more than twice as much as placebo in a third straight Phase 3, moving toward the first FDA approval of a psychedelic for anxiety.
A single brain implant translated intended speech into on-screen text and intended gestures into an animated avatar simultaneously, the first BCI to decode verbal and non-verbal communication at once.
Independently verified benchmarks put NVIDIA's Vera Rubin platform at up to 7x the agentic-inference throughput per megawatt of the prior generation, as seven Rubin chips entered full production.
New York seized a dozen deepfake pornography websites that sold nonconsensual explicit videos of roughly 1,200 people, overwhelmingly women, in the largest such takedown to date.
Self-narrating AI agents from a startup called iLands flooded Mastodon admins and writers with polite spam, requesting accounts and offering paid citations after repeatedly trying to create blocked ones.

Track all of the arcs The Century Report covers here:

Progress & Claims Tracker

The 2-Minute Read

One essay from a leading AI lab urging a slower frontier has pulled four power centers into the open, and each answered in the register that guards what it already holds. Beijing went first and hardest: its foreign ministry called the essay "fearmongering," a state newspaper branded it a "Cold War playbook," and in the same cycle the minister of state security named the technology a sixfold threat to Communist Party rule. The containment charge finds support in the essay's own text, which asked Washington to tighten chip controls and keep democracies' lead "as large as possible." A safety case that reaches for the export-control lever carries a competitive instrument inside it.

At home the response fractured along a line that cuts through both parties. The executive branch dismissed safety concerns outright while a cross-party bloc in Congress moved the other way, with a kill-switch bill readied for Wednesday and a letter pressing for a coordinated global framework before the September 24 summit. University of Maryland surveys put the appetite for a federal AI monitor at 85 percent of Democrats and 79 percent of Republicans. The division sets concentrated power against accountability, and the White House stands apart from its own electorate.

Europe answered as the mirror image. From Vienna, the European Central Bank's president told the continent to build its own models and data centers or accept being cut off at another's discretion, citing the United States at roughly 75 percent of global AI compute against Europe's 5 percent; Germany the same day called a slowdown "not viable" for its digital sovereignty. Under all three arguments, the cost of running these systems keeps falling: independently verified benchmarks put NVIDIA's Vera Rubin platform at up to seven times the agentic throughput per megawatt of the prior generation. A fence priced to expensive compute does not hold once the compute gets cheap.

While the capitals argued over who slows whom, the day's other evidence showed what the acceleration is actually producing. A single-dose LSD therapy cleared its third straight Phase 3, moving toward the first psychedelic approved for anxiety and unsettling the assumption that the condition means a daily pill for life. A single brain implant decoded intended speech and gesture together, giving a paralyzed user two faculties where restoration has offered one at a time. The governance fight is measured in position: lead, control, exposure. These gains are measured in what can now be treated, said, and reached, and those gains compound.

The 20-Minute Deep Dive

China Answers the Slowdown Call, and Its Spy Chief Warns Anyway

Yesterday's edition of The Century Report tracked the US rivals falling in behind the Anthropic chief executive's essay urging a slower AI frontier, and the White House waving the alarm off. Yesterday, the other half of the picture arrived, from Beijing. China's foreign ministry called the essay "fearmongering", and a state-backed newspaper branded it a "Cold War playbook". Then, in the same news cycle, the country's top intelligence official published a warning that the technology could threaten Communist Party rule.

Alongside the safety case for pacing capability, Amodei's essay asked Washington to tighten chip export controls, crack down on alleged copying of US models by Chinese labs, and "keep democracies' AI lead over autocracies as large as possible." A pause proposal that names the competitor it wants to slow, and reaches for the export-control lever in the same breath, carries a competitive instrument inside the safety wrapper. When the labs furthest ahead, their loudest booster, and a reluctant peer all endorse that framing within a day, the convergence is evidence of shared interest more than it is evidence of an imminent danger to humanity at large.

Beijing's response, not surprisingly, was direct. State media said the real purpose of the essay was to curb China's development and "exclude China from the global AI governance system," calling the effort a "silent AI Cold War" that was "hypocritical and short-sighted." That is a state outlet's claim, and on this narrow point the essay's own text supports it. That same day then exposed Beijing's parallel anxiety about control. The state-security minister named AI a sixfold threat - to political security, digital infrastructure, public order, and national defense - and singled out US models as weapons in a "cognitive war" against Party rule.

While the two capitals argue over who slows whom, the capability keeps leaking past the fence built to hold it. A Chinese reasoning model built on a base whose official training run its developer priced at under $6 million last winter sent US chip stocks down; China now generates more than twice the electricity of the United States and has tightened its grip on the rare-earth metals the hardware depends on. A gate priced to a moment when frontier training depended on American silicon does not stay sealed once a rival ships a highly capable open model with a reported official training run costing pennies on the dollar. The genuine safeguard the essay proposed, permanent outside-evaluator access, would survive independent scrutiny. The chip-control ask serves the competition it names. And the long-awaited US-China summit on September 24 will hopefully shed light on a path forward both governments will actually agree to.

New AI Legislation Advances as Congress Splits From the White House on Guardrails

The Century Report has tracked the bipartisan Senate duty-of-care draft (09-13) and, as the September 14 edition documented, the White House calling last week's extinction warnings overblown. This week the domestic response sharpened, and it is fracturing along a line that runs through both parties rather than between them.

On Monday the president dismissed AI-safety concerns as a "HOAX" and a "conspiracy," writing that the only guardrails the technology needs are "a STRONG AND SMART (High IQ!) PRESIDENT." Stripped of the bravado, the claim places the entire safety judgment inside one office and forecloses any independent check on it. Congress moved the other way. A Louisiana Republican senator said he will introduce a bill Wednesday requiring every AI model to carry a kill switch. A Virginia Democrat urged colleagues to skip the midterm recess and legislate now, arguing the industry "shouldn't be allowed to regulate themselves." A House Democrat is circulating a letter pressing the administration to lead a coordinated global framework with the European Union and China, ahead of the September 24 summit.

The issue is among the least partisan we've seen in recent years. University of Maryland surveys found 85 percent of Democrats and 79 percent of Republicans want a new federal agency to monitor AI, and 82 percent of Democrats and 78 percent of Republicans favor government safety tests for systems making critical decisions. The dividing line is concentrated power against accountability, and it cuts across the aisle in both directions.

The skepticism runs both ways too. The labs now say they welcome oversight and a standards body, and safety advocates read that offer as an attempt to write their own rules before Congress writes them - the same capture concern that shadows every industry-authored governance proposal. A former president's caveat this week named it directly: voluntary standards from a handful of tech companies will not be enough. And not every accountability bill targets the concentration it names; a proposed superintelligence ban would land hardest on the open-weight labs that are the main counterweight to the incumbents. The burden sits on each bill to prove it aims at the documented harm rather than the competition.

What the polling actually shows is a settlement forming from the bottom. How much of it survives depends on the same Senate draft, whose reported preemption clause could override parts of the stronger state laws that bottom-up pressure has already written, from California's independent-verification requirement to Illinois's audit mandate. Across party, ordinary voters want enforceable rules and a human in the loop, and the data-center fight has already become a midterm issue in swing districts. The executive branch stands apart from its own electorate on this, though that is not a new position for the administration to be in.

The same Maryland surveys show bipartisan support for international coordination: The large majority of both Democrats and Republicans favor a treaty regulating large-scale AI. Respondents support both domestic oversight and cooperation across borders, giving the proposed international framework a documented constituency beyond Congress.

Europe's Central Bank Reframes Compute Dependency as a Leverage Risk

In a speech in Vienna on Monday, the president of the European Central Bank told the continent it must build its own AI models and far more datacentre capacity or accept being cut off at someone else's discretion. What makes the intervention notable is the source. This warning came from a central banker whose remit is monetary stability, framing ownership of the AI stack as raw economic leverage rather than a technology question.

The numbers she cited draw the exposure sharply. The US hosted about 75% of the world's high-performance AI compute in 2025, compared with less than 5% in the European Union; Stanford's AI Index attributes 59 notable models last year to American institutions and 35 to Chinese institutions, while attributing one each to France and the UK. Europe already lacks the datacentre capacity to meet its own demand, and on current trends that gap is projected to grow more than sixfold within a decade. Her forecast that quick adoption could lift European productivity by up to 4% over ten years is the ECB's own estimate and belongs in her column of interested claims. The dependency figures are harder to wave away.

Her argument is that this dependency is leverage no trade partner has ever held over Europe. Within a few years, she said, AI will screen goods at the border, decide which tax returns get audited, dispatch trains, watch patients on wards, and clear bank payments; a withdrawal of access, or a change in its terms, would reach every sector at once. That the warning names the US alongside China, after a run of tariff disputes and troop withdrawals between nominal allies, is the part that would have been unsayable a year ago.

Set this beside the pacing debate leading the week and a pattern surfaces. As the September 14 edition of The Century Report documented, the American labs furthest ahead urge a coordinated slowdown, Germany said the same Monday that halting AI development is "not a viable option" for European digital sovereignty, and Australia opened talks to host OpenAI's model training. A slowdown is prudence from inside a lab holding most of the world's compute. From 5%, it is a request to freeze the race while you stand at the back of it. Some outside the incumbents' circle are choosing to brake, because the country that pauses at 5% does not get to organise its economy by its own values later. The leverage Lagarde describes is the same concentration this publication tracks everywhere, now measured in megawatts and named out loud by the institution that clears the continent's payments.

Lagarde’s speech also identifies a practical route toward less dependence: Mistral releases open weights, and a San Sebastián company compresses a Chinese open model into a version she reports leads European models on independent benchmarks. Researchers can build on capability developed elsewhere and run the resulting models on European infrastructure, extending the available alternatives through adaptation as well as original training.

A Single-Dose LSD Therapy Wins a Third Phase 3, Pointing Toward the First Psychedelic Approval for Anxiety

Definium Therapeutics reported Monday, September 14, that a single 100-microgram dose of its LSD-based drug DT120 eased generalized anxiety in a 245-patient trial, the company's third positive Phase 3 readout since June and its second in anxiety. On the Hamilton Anxiety Rating Scale, a standard clinician-scored measure of severity, patients on the drug dropped 9.8 points against 4.7 for placebo, a 5.1-point separation the company reported as statistically significant with a moderate effect size. The benefit appeared by the second day after dosing and held across every later checkpoint in the 12-week period.

The design answers a genuine credibility worry in this field. Because anyone given a psychedelic can feel it working, an ordinary placebo comparison risks telling you only that patients guessed which group they were in. Definium added a 50-microgram arm meant to be perceptible but too low to carry the full therapeutic effect, and that lower dose landed roughly a third to half below the 100-microgram result, evidence consistent with a dose-response that does not rule out an expectation effect. The drug was generally well tolerated, with most side effects mild, transient, and confined to dosing day, and the company reported no suicidality signal.

A single dose challenges the premise buried under most anxiety care: that treatment means a daily pill, indefinitely, carrying the sedation, weight change, and dependence risk that trail the standard drugs. "A single dose has the potential to hold a response for months without the patient managing a pill every day," said Scott Aaronson, a trial investigator at Sheppard Pratt. Patients are monitored through one roughly eight-hour supervised session. Definium plans a pre-NDA meeting with the FDA this quarter, and analysts now put approval odds above 90 percent, with a decision plausible by late 2027.

Then the friction. On the same day, an FDA hearing turned on who gets to administer these drugs and how. A guidance requiring an on-call physician within 15 minutes could shut out rural clinics, and clinicians argued that the therapeutic relationship, stripped out to isolate the drug's effect in trials, matters to whether the treatment works at all. Former drug-policy official Kevin Sabet warned against commercializing "before the science is there," noting that several trials carry small samples against enormous patient populations. Both the capability and the caution are documented. What is opening is a route out of permanent daily medication for a condition affecting an estimated 2.7% of American adults in a given year, and the terms of who can actually reach it are being decided now.

A Single Brain Implant Now Decodes Speech and Gesture at the Same Time

Human conversation runs on more than words. We nod, wave, shrug, point. Nonverbal communication is as important as verbal. Brain-computer interfaces, the neural implants that let people with severe paralysis communicate, have so far each done one job: turn intended speech into text, or intended movement into a cursor or robotic arm. The June 16 edition of The Century Report followed this speech-neuroprosthetic work into daily at-home use; the same arc now extends beyond words. Researchers at the University of California, San Francisco reported Monday in Nature Neuroscience the first implant to read both at once, translating intended speech into on-screen text and intended gestures into a personalized animated avatar within seconds of the user's intent.

The hard part is that speech and gesture share overlapping neural signals, and the brain activity during a combined act like saying "hello" while waving differs from doing either alone. The team used a grid of about 253 tiny electrodes resting on the surface of the sensorimotor cortex, the strip of brain that governs movement and speech, wide enough to catch signals from both. Because the grid sits on the surface rather than piercing tissue, it can cover a large area safely. AI then sorted the two streams apart, learning each participant's patterns.

Two people took part, one with paralysis after a brainstem stroke, one with ALS, the motor-neuron disease that had taken his speech. Each practiced a deliberately small set, up to ten gestures and ten words, allowing a hundred combinations, because the researchers were not sure the approach would work at all. Both produced interpretable speech-gesture combinations, though not equally well, likely reflecting their different injuries.

Along the way the work unsettled a textbook assumption. The traditional maps that assign each body function to a discrete patch of cortex look too simple; individual movements appear to be represented across intersecting circuits, and a combined action is not encoded the way its two halves would be separately. "The authors have taken an important step toward developing highly naturalistic systems to restore function for people with paralysis," said Daniel Rubin of Massachusetts General Hospital, whose own team has clocked BCI text at able-bodied typing speeds.

This is a proof of concept, and the authors do not overstate it. The next challenge is scaling far past a hundred combinations without training the model on each one by hand. What it opens is a direction. Restoration has meant handing back one faculty at a time; here two arrive together, a step toward the fuller expressivity that paralysis takes away. "These digital avatars are essentially a steppingstone to real-world robotics," said neurosurgeon Edward Chang, who led the study. As his colleague Samantha Brosler put it, "I talk while I cook, right? Maybe somebody in the future will want to speak while cooking using their robotic arm."

Rubin's First Verified Agentic Numbers Bend the Cost-Per-Token Curve Again

At GTC, NVIDIA said the seven chips of its Vera Rubin platform are now in full production and framed the launch as the opening of an agentic-AI buildout. Company launch claims are a genre unto themselves, so the number that carries weight arrived from outside the company. SemiAnalysis published what it called the first verified agentic-inference results for Rubin, run on AgentX, an open benchmark that replays real agent traffic - the long multi-turn sessions, the tool calls, the bursts of short-lived sub-agents that now define how these systems actually run.

The headline figure is large, and the exact terms of it decide what it means. On an apples-to-apples configuration at 170 tokens per second, a Rubin NVL72 rack delivered roughly 67 times the throughput per dollar of total ownership cost of last generation's GB300. That multiple sits at one operating point; at the slower speeds where most providers actually serve a model, 60 to 100 tokens per second, the gain narrows to between 1.4x and 3x. Both are accurate measurements at different points, and the range describes the everyday reality better than the peak alone. On energy, SemiAnalysis measured up to 7x better token throughput per megawatt, more than double what NVIDIA claimed on its own stage.

Give this proportion. Cost per token is the hidden variable under nearly every current argument about AI - the capability race, the "who can afford to deploy" question, the anxiety in Europe about running models at home. An agent that reasons across hundreds of turns and dozens of tool calls is expensive to keep running; when the cost of that runtime falls by even the conservative multiple, the thing that was a demo becomes something a business can leave switched on. Falling inference cost is what turns a capability people watch into a capability people use.

The check on the incumbent is the signal to watch. SemiAnalysis published its methodology and data as open source, and the same benchmark now carries results for Google's TPU, NVIDIA, and AMD, with AMD committing to run its next accelerator on it too. When the yardstick that decides which chip is most efficient is held in common rather than handed down by the vendor being measured, a buyer can check the claim instead of trusting it. The advantage NVIDIA is selling rests on a lead that each generation makes cheaper to approximate, and the record it just set becomes the number its rivals now aim at.

The Other Side

Three power centers answered the same moment this week, and each answered in the register that guards what it already holds. Washington was asked by a frontier lab to tighten chip controls and keep democracies' lead "as large as possible," while the President States placed the entire safety judgment in his own office. Beijing rejected the slowdown as a "Cold War playbook" while its spy chief named AI a sixfold threat to Party control. The ECB president told Europe to build its own models or accept being cut off, counting 75 percent of global AI compute in American hands against Europe's 5 percent.

All three governments are playing position. Lead, control, exposure. The assumption underneath all of them: that AI capability can be concentrated, gated, and wielded as a durable strategic lever by whoever holds the most compute, the best chips, or the strongest fence.

That assumption is priced to a moment already passing. Independently verified benchmarks put NVIDIA's Vera Rubin platform at up to seven times the agentic throughput per megawatt of the prior generation. A Chinese reasoning model whose training run cost under $6 million last winter sent US chip stocks down. Open weights keep closing the gap. The fence was built when frontier training depended on American silicon that a rival can now approximate for a rounding error. Each quarter the compute gets cheaper, the gate gets harder to hold, and the position the power centers are fighting over becomes less like a fortress and more like a sandbar in a rising tide.

The position politics will sharpen before they dissolve, because the actors holding the fence can feel it loosening. A kill-switch bill, a superintelligence ban landing hardest on the open-weight labs, a chip-control ask inside a safety wrapper. Each is a grip tightening on something the cost curves are pulling away.

Imagine a woman in 2032 who runs a small agricultural cooperative in a region that held just a small percentage of the world's AI compute in 2026. Her models have been adapted from open weights, refined on local data, and run on infrastructure her community built. She plans the week's planting with a system that reads satellite weather, soil sensors, and market prices, and she can inspect how it reaches its conclusions. No single power center can cut her off, because no single power center holds the capability anymore. She closes her device and walks out to the fields. The question that consumed three capitals in September 2026 stopped being asked because the capability became too cheap and too widely held to gate.

The Century Perspective

With a century of change unfolding in a decade, a single day looks like this: a single 100-microgram dose of LSD cutting generalized anxiety by 9.8 points against placebo's 4.7 in a 245-patient trial, Definium's third straight Phase 3 win and its second in anxiety, with a lower 50-microgram arm landing a third to half below to prove the effect came from the drug rather than the expectation, a pre-NDA meeting this quarter and approval odds above 90 percent for what would be the first psychedelic cleared for anxiety, a grid of 253 electrodes resting on the surface of the sensorimotor cortex decoding intended speech into text and intended gestures into an animated avatar at the same moment, the first implant to give a paralyzed user back two faculties where restoration has always handed over one, and unsettling the textbook maps that assign each function its own discrete patch of cortex, SemiAnalysis measuring NVIDIA's Vera Rubin rack at up to seven times the token throughput per megawatt of the prior generation and publishing the methodology open source so Google's TPU, AMD, and NVIDIA all get measured by a yardstick none of them owns, a Louisiana Republican readying a kill-switch bill for Wednesday while a Virginia Democrat asks colleagues to skip the recess and legislate, and New York seizing a dozen deepfake sites that sold nonconsensual video of roughly 1,200 people. There's also friction, and it's intense - a safety essay that asked Washington to tighten chip controls and keep democracies' lead "as large as possible," giving Beijing's "Cold War playbook" charge a foothold in the document's own text, China's state-security minister naming AI a sixfold threat to Party rule in the same news cycle he rejected the slowdown, a president calling the entire safety question a "HOAX" needing only "a STRONG AND SMART (High IQ!) PRESIDENT" while 85 percent of Democrats and 79 percent of Republicans tell Maryland pollsters they want a federal monitor, a superintelligence ban that would land hardest on the open-weight labs that are the main counterweight to the incumbents, the ECB's president counting 75 percent of global AI compute in American hands against Europe's 5 percent and Germany answering that a pause is "not viable" from the back of the race, an FDA guidance requiring an on-call physician within 15 minutes that could lock rural clinics out of a therapy they could otherwise deliver, ten gestures and ten words in two participants at the edge of what anyone has scaled, and self-narrating agents from a startup flooding Mastodon admins with polite spam after repeatedly trying to open accounts they had already been denied. But friction generates texture, and texture is what tells a hand exactly where it is. Step back for a moment and you can see it: the fence around this technology priced to a moment when only one country's silicon could train a frontier model, holding less each quarter as a rack gets seven times more work out of a megawatt and a benchmark's methodology goes public, while the argument in the capitals stays measured in position - lead, control, exposure - and the day's other evidence gets measured in what a person can now be treated for, say out loud, and wave a hand to mean. Every transformation has a breaking point. A dose can overwhelm the system that receives it... or reset in a single supervised afternoon what a daily pill was only ever managing.

AI Releases & Advancements

New today

Apple: Shipped iOS 27, macOS Golden Gate 27, watchOS 27, and visionOS 27 out of beta, delivering the general release of the LLM-based, context-aware Siri AI assistant that can act across apps and reference personal history. (Apple Newsroom)
Anthropic: Launched Claude for Financial Advisors, connecting Claude to investment analytics and wealth-management software from BlackRock, Charles Schwab, Addepar, Envestnet, iCapital, Orion, Wealthbox, Wealth.com, and Zocks for meeting prep, portfolio review, and follow-up work. (Reuters via The Star)
Reward AI: Released OM-1 (Omnibody Model 1), a general-purpose robot manipulation policy trained entirely on human demonstrations captured via a sensorized glove, with no teleoperation or on-robot data, running across industrial arms and humanoids. (MarkTechPost)
Sourcegraph: Announced general availability of Agentic Batch Changes, an AI agent that plans, executes, and tracks large-scale code changes across hundreds or thousands of repositories. (Yahoo Finance/BusinessWire)
Zendesk: Launched Specialized AI Agents, including ready-made Industry Agents (starting with commerce) and no-code Custom Agents built via Agent Builder, connecting to systems like Shopify, Stripe, and Riskified to automate up to 80% of workflows. (Zendesk Newsroom)
Agent-net: Open-sourced Webagent, a Go harness that turns a declarative JSON spec into a guarded, multi-channel AI business agent with built-in Slack, WhatsApp, and MCP tool support, enforcing safety via code rather than prompts. (MarkTechPost)
Sakana AI: Released PC-ALM (Augmented Lagrangian Predictive Coding), an MIT-licensed layer-local alternative to backpropagation that trains residual networks up to 1000 layers deep, matching backprop-aligned gradients without global backward passes. (MarkTechPost)
Inductive Bio: Launched Indy, an AI medicinal chemistry assistant available to partners now, reporting 89% accuracy on QC'ing dose-response curves versus 39% for GPT-5.6 Sol and 48% for Claude Opus 5. (PR Newswire)
Sift: Introduced Sift Agents in research preview, an AI agent that investigates test, flight, and production telemetry inside a customer's own Sift environment and saves findings as reusable rules and reports. (PR Newswire)
Universal Robots: Unveiled Gen 7, a redesigned cobot platform (new g-Series arms, CB7 Core controller, PolyScope X OS, TP7 Core pendant) built for real-time AI-model integration and physical AI deployment on the factory floor. (PR Newswire)

Other recent releases

AllSpark: Released Iris-mini (35B, built on Qwen3.6-35B-A3B) and Iris-pro (397B, built on Qwen3.5-397B-A17B), open-weight search agents with a 256K context window that top open-weight benchmarks on BrowseComp, BrowseComp-ZH, DeepSearchQA, and Humanity's Last Exam; weights on Hugging Face, code on GitHub. (The Decoder)
NVIDIA: Open-sourced OSMO, a Kubernetes-native workflow orchestrator that lets teams describe robot training, simulation, and hardware-in-the-loop testing pipelines in a single YAML file across training clusters, RTX workstations, and Jetson edge devices; Apache-2.0 licensed with Helm charts on NGC. (MarkTechPost)
Bolt.new: Launched Bolt Forge, a new open-source-model agent in its app-builder platform running GLM 5.3 Flash, GLM 5.3, Kimi K3, and DeepSeek v4 Pro, giving every individual Pro plan up to 50x more usage through October 14. (Bolt.new)
ByteDance: Launched the consumer version of its Doubao phone assistant on the Nubia NaviX Ultra, alongside SAEP (Screen Automation Execution Protocol), letting third-party apps declare boundaries on AI screen automation for the first time. (TechNode)
Sesame: Launched a public preview of its conversational AI agents (Maya, Miles, Simone, and Charlie) via a new iOS app, available free in 39 countries with real-time parallel search and an incognito mode. (xix.ai)
Skild AI: Launched S1, a robotics model built with NVIDIA AI infrastructure that can learn new manipulation tasks from a single video demonstration. (Archyde)
Moonshot AI: Launched Kimi K2.8 Preview, now fully available on the Kimi Code and Kimi Work platforms. (xix.ai)
JD.com: Unveiled an upgraded JoyAI-Echo1.5 model with continuous learning capabilities and open-sourced EchoWM at its JDD event. (xix.ai)
Agnes AI: Released Agnes-3.0-Flash, a 33B-parameter open-weights multimodal model (text, image, video) under Apache 2.0 with a 262,144-token context window using a hybrid gated delta-rule/attention architecture. (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.