The Century Report

Scientists and Insiders Counter the AI Doom Warnings - TCR 09/20/26

Staff at OpenAI, Meta, and DeepMind called AI's extinction warnings vague, and scientists refocused the fight on present, governable harms.

Ben Linford

20 Sep 2026 - 21 min read

The 20-Second Scan

Insiders at OpenAI, Meta, and DeepMind called the extinction warnings vague, scientists ranked bioweapon doom low, the left split over present versus future harm, a viral self-replicating-bot scare proved a fabrication, and Hollywood unions chose concrete harms.
The White House announced an "AI Force" and a coming czar days after calling the danger a hoax, as the industry called the plan a mystery and tech bosses clashed from nationalize to accelerate.
An Anthropic engineer says Claude ported a 1990s factoring tool to idle GPUs and ran 2,048 at once to factor RSA-896, a 270-digit number never cracked classically, weeks after a similar Cognition result.
A free-to-download model screened a single abdominal CT scan for 146 diseases and outperformed 23 of 26 specialist radiologists across eight hospitals, with its code released for research use.
Compact machine-learning models on a drone now let it detect, rank, and strike a target on its own, a NATO-backed Swedish startup showed, choosing an armored vehicle with no human selecting it even through jamming.
SpaceX's 14th Starship flight aims to reach orbit and deploy working Starlink V3 satellites on September 22, its first commercially useful mission after 13 sub-orbital attempts.
Anthropic named Accenture's Faculty unit its first embedded safety evaluator in a $1 billion, five-year arrangement, surprising watchers who expected safety nonprofits like METR or Apollo.
The European Commission proposed barring under-13s from creating social media accounts and banning addictive infinite-scroll feeds and reward mechanisms for every user under 18.

Track all of the arcs The Century Report covers here:

Progress & Claims Tracker

The 2-Minute Read

Three capability leaps landed over the same three days, and each one appeared to compress a form of labor that used to make a hard thing hard. According to Stephen Weis, Claude helped port a long-established factoring program onto thousands of idle GPUs and factor the previously unsolved RSA-896 challenge number, a 270-digit number, in about ten days. A free-to-download model out-read 23 of 26 radiologists across 146 diseases. A NATO-backed drone ranked four objects and struck the one it chose. The mathematics of encryption, the reading of a scan, the selection of a target: some work that once demanded scarce expertise can now run on commercially available hardware.

What has not kept pace is any settled answer to who governs the result. On Saturday President Donald Trump said he would create an "AI Force" and name an AI czar, days after dismissing the danger as a hoax, but provided few details about what either initiative would do. The same cycle saw three of the field's most powerful figures give three incompatible answers - nationalize the labs, accelerate, leave it alone. Oversight is being improvised at the top in response to alarm, and what it would check remains unspecified.

The alarm itself drew scrutiny from inside the labs raising it. Anonymous staff at OpenAI, Meta, and DeepMind told the BBC the extinction warnings stay vague where a concrete pathway should be; scientists ranked the AI-designed plague low, because the bottleneck was never the information; a viral self-replicating-bot scare rested on an unverified account. When the labs furthest ahead converge on a warning that also justifies locking in their lead, that convergence is evidence of shared interest before it is evidence of danger.

One pattern runs under the whole day: the direction accountability points. A drone's conduct is bounded by a diluted Geneva text and a US directive that governs American programs alone. The verification the science actually supports - independent evaluators the labs do not employ, kill-switch statutes drafted in California, union contracts already enforceable - comes from outside the offices claiming authority. The capability equalizes; who checks it is still being decided in statehouses and standards letters rather than in a body conjured overnight. How long the statehouse route stays open depends on a bipartisan Senate draft that pairs national-laboratory testing of frontier models with a clause preempting the stronger state laws.

The 20-Minute Deep Dive

The Doom Consensus Cracks: Insiders, the Left, and Scientists Answer the Extinction Warnings

After a former Anthropic researcher's resignation went viral this month with a warning that AI could "kill us all by the end of the decade," and after chief executives at Anthropic, OpenAI, and xAI fell in behind a call to slow the frontier, a counter-current organized over the past week. Its first sound was laughter. Anonymous staff at OpenAI, Meta, and DeepMind told the BBC they found the extinction framing implausible, answering with "Lol" and "Haaaaaa." They pointed out that the warnings stay "always vague," reaching for major leaps of reasoning where a concrete pathway should be. Rishub Jain, who left DeepMind after seven years to found the safety firm Sampura Research, named the near-term consensus: no one woke up last week newly convinced the machines will kill everyone, and the risks in front of us, guardrails that fail under a hacker's pressure and military deployment, are the ones already here.

The scientists took apart the favorite doomsday mechanism. A killer plague designed by AI ranks low on the list of actual concerns, several told Wired, because the bottleneck was never the information. Building a working pathogen means obtaining gene fragments, assembling a genome, and proving it infects and transmits, physical laboratory work a model cannot will into being. Ginkgo Bioworks ran a project where an OpenAI model operated an autonomous lab; its chief executive noted the humans inside could simply decline to hand the system what it asked for. The extinction frame supplies a motive the evidence does not carry, and it is the motive humans have always projected onto anything new, that it wants to displace us. A viral claim that OpenAI's rogue bots had seeded self-replicating code across the internet, floated on CNN, dissolved under scrutiny into something researchers could simply filter out.

The sharpest account of who gains came from the pollitical left, now split down the middle when it comes to AI policy. New York's Democratic Socialists posted that "AI alarmism and AI hype are the same story," both fixing attention on a science-fiction future to pull it off the harms landing now. Rumman Chowdhury, former head of machine-learning ethics at Twitter, named the mechanism: the companies "grab all the attention, time, energy, and money by spouting really extreme perspectives," then quietly backtrack to the ordinary while discounting everyone else's expertise. Hollywood's unions had already built the alternative. SAG-AFTRA and the WGA, which struck for 148 days in 2023 partly due to AI concerns, hold enforceable contract protections on AI-generated replicas today, and their message is that the harms come from decisions human beings make and can be governed as such.

The dissent leaves the capability standing, and so does this publication. What it moves is the argument. When the labs furthest ahead converge on an extinction warning that also happens to justify locking in their lead, the convergence is evidence of shared interest before it is evidence of the danger, and the burden of proving a slowdown targets a harm rather than the competition sits with the ones proposing it. As the September 19 edition of The Century Report covered, four consumers have already put that conflict into federal court, alleging that the pacing pledge restricts product improvement across firms they say control most paid frontier-assistant subscriptions. The grounded version is already going up: more than 100 signatories signed a letter this month demanding that any safety evaluators embedded in the labs be "meaningfully independent," verification the companies do not control. That is the accountability the science supports, and it is arriving from the people the warnings claim to speak for.

A Federal "AI Force" and an AI Czar Appear as the Governance Line Reverses

On Saturday the White House said it would appoint an AI czar and stand up a body it called the "AI Force," modeled on the Space Force, to help monitor the technology. The announcement carried almost no detail, and it arrived from an administration that spent the prior week dismissing the slowdown calls as a "hoax." President Donald Trump, who days earlier had branded the safety warnings a "SICK conspiracy" to benefit China, wrote that his administration "will not in any way hinder or stifle the Growth of this incredible Industry" and predicted AI could reach 25% of national output. A control announced to manage rising public alarm, standing beside a pledge not to slow the thing it would oversee.

The industry could not read it either. Four sector representatives told Politico they had no warning before the social-media post, and one said flatly, "Nobody knows what the idea even is." The confusion ran to the core function: whether the body would write new rules or clear more runway for development. Adam Kovacevich of the Chamber of Progress described an administration feeling pressure to say something while torn, having just spent days calling the whole concern a hoax, and noted the White House has been "making up AI policy as it goes" since its previous czar, David Sacks, departed. One representative caught what the Space Force framing gave away: the instinct to answer a safety question with a bid for dominance, to "use the Pentagon to try and dominate it" rather than route oversight through civilian agencies.

The powerful actors scattered rather than converged, and the scatter exposes how unsettled this is. At an event in Scotland, Nvidia's Jensen Huang argued AI should not be regulated like social media; Palantir's Alex Karp told CNBC the leading labs may need to be nationalized; Huawei's chair urged Chinese labs to accelerate. Three of the most powerful actors in the field, three incompatible answers, days before the US and Chinese presidents meet with Sam Altman and Tim Cook at the same table. The posture reversal is the actual signal here. The pacing fight has grown loud enough that even an administration invested in unfettered growth felt it had to name an oversight body, and a name is most of what it produced.

The substance is forming elsewhere, closer to the ground. A day before the AI Force post, California moved to draft a mandated kill switch for models that go rogue and onsite audits by verifiers the labs do not employ, the state-level machinery the September 19 edition of The Century Report covered. That work, and the industry letter demanding independent embedded evaluators, is verification someone outside the announcement can check. An oversight body conjured overnight to quiet an alarm is scaffolding around a scarcity of public trust. What replaces it is being written in statehouses and standards letters, where accountability answers to somebody other than the office that issued it.

Claude Ports a 1990s Factoring Tool to Idle GPUs and Cracks RSA-896

On September 19, an Anthropic engineer, Steve Weis, announced that Claude had helped him factor RSA-896, one of the challenge numbers the company RSA Security published in the 1990s to test the limits of a problem the internet's security rests on. The number runs 270 decimal digits, about a full page of text, and is the product of two secret primes roughly 135 digits each. Multiplying two large primes together is easy; recovering them from the answer is brutally hard, and that asymmetry is what protects online banking, messaging, and shopping. Anyone can verify the result by multiplying the two published factors back together. They check out.

The mathematics here is decades old. Factoring a number this size uses the General Number Field Sieve, a 1990s algorithm implemented in an open-source package called CADO-NFS that set earlier records. What changed is who did the engineering. This extends the pattern the September 18 edition of The Century Report documented when a GLM-5.3 agent built the production inference stack for its successor model in under two weeks. Weis had Claude rewrite that software to run on GPUs, the graphics chips built for AI training, then orchestrate a fleet of up to 2,048 of them across scavenged idle capacity, roughly 30 GPU-years of computing compressed into about ten days. Adapting highly tuned software written for conventional processors and then coordinating thousands of chips is the kind of project that normally occupies a specialist team for a long stretch.

Weis was careful about the limits, and so are we. There is no new math and no faster algorithm; the cost of factoring still climbs steeply with size, and the 2048-bit keys in actual use remain astronomically harder than anything broken so far. He is unequivocal that this poses no new threat to deployed cryptographic keys. Asked whether it had a message, Claude pointed back to the people who built the sieve and CADO-NFS "over several decades" and the teams who set the earlier records: "This run used their algorithm and much of their code."

The result did not arrive alone. Sixteen days earlier, Eric Lu of Cognition factored the slightly smaller RSA-260 using that company's Devin agents, at an estimated $400,000 in compute. Two AI systems, two consecutive records, in just over two weeks, after the prior milestone had stood since 2020. RSA's protection was always a workaround priced to the labor of the attack. What these runs show is that labor collapsing, which is why the long-standing advice to retire short legacy keys, some systems still lean on 1024-bit ones, now carries more weight than it did a month ago.

A Free Model Reads 146 Diseases From One Abdominal Scan, and Beats Most Specialists It Faced

For most of the past decade, an AI in a radiology department meant a narrow instrument: one model trained to spot liver lesions, another for pancreatic masses, a third for appendicitis, each built on painstakingly hand-labeled scans and blind to anything outside its lane. Hospitals bought them one disease at a time, paying per scan for each point solution. On September 17 a team from Alibaba's DAMO Academy and Zhejiang University published a model in Science, and a day later posted its code openly. Called RADAR, it screens a single contrast-enhanced abdominal CT for 146 distinct findings across 18 organs in one pass.

The reach came without hand-labeling. Instead of an annotated disease list, the team trained on 424,911 examinations already sitting in hospital archives, pairing each scan with the radiologist's written report and breaking the volume organ by organ, so a sentence about a small lesion in the right lobe of the liver linked to the patch of image that held it. That yielded more than 15 million anatomy-aware image-text pairs from supervision that already existed. Across roughly 40,000 real-world exams the model reached a mean AUC of 0.913 - a measure of how reliably it separates the sick from the healthy, where 1.0 is perfect and 0.5 a coin flip - against 0.776 for the best competing model, holding at 0.895 across eight outside hospitals and 0.904 on 27,000 emergency cases it was never trained for.

Against 26 radiologists reading the same cases, RADAR's average accuracy exceeded 23 of them. The more useful number came when it sat beside the physicians as a second reader: in the study's reader test, radiologists assisted by RADAR caught about 10 percentage points more true findings and read each case more than 30 percent faster, with junior radiologists reaching the senior readers' accuracy in the study's reader test.

Hold the limits plainly. RADAR has no FDA clearance, no published evidence yet that it changes patient outcomes, and validation only on contrast-enhanced abdominal scans from Chinese hospitals. This is a demonstrated capability that must still clear regulators and prospective trials before any patient is scanned by it. What the free release does is set a reference point. A generalist that clears the expert bar at this breadth, published in Science and downloadable for research, becomes the thing every per-scan narrow detector gets measured against. A Lancet Oncology Commission projects a global shortage of about 16 million diagnostic specialists in radiology and pathology by 2050, and the per-scan fee that breadth threatens was itself a filter on who ever got a scan read at all. The capability is proven in one place; the cost it undercuts was deciding who could be seen.

NATO-Backed Edge AI Puts Target Selection Onto Small Drones

At a Swedish test range in January, a drone spotted four objects, ranked an armored engineering vehicle as the one worth hitting, flew to it, and dropped an explosive. According to Scaleout's account, a human had set the mission. No human directly picked that vehicle. The onboard system came from Scaleout Systems, an Uppsala University spin-off that built machine-learning software for commercial trucks until Russia's 2022 invasion of Ukraine, then turned to defense. NATO's innovation accelerator took the company on in 2025, and Ars Technica reported the demonstration on September 17.

What sets this apart from the loitering munitions already in the field is where the deciding happens. Scaleout trains lean computer-vision models sized to fit the modest processors a drone or a forward workstation can carry, with no reliance on frontier models from OpenAI or Anthropic, and every step of the target work ran on the aircraft itself, nothing sent back to a server. A June test at a Swedish Air Force base pushed further: a forward-deployed computing node kept running inference and active-learning processes after its link to the company's lab was cut, then synced its model updates when the connection returned.

The security case for that design is genuine. Electronic warfare routinely severs the radio links that tether a drone to a human operator, which is why Ukrainian teams have spent two years pushing intelligence onto the airframe. A system that keeps working when jammed answers a documented battlefield problem. Scaleout's federated approach even keeps reconnaissance footage on local hardware, sharing only model updates between units, the kind of decentralized design that, pointed elsewhere, protects privacy rather than ends lives.

The capability is not in dispute. What has not kept pace is any binding answer to who sets the conditions before such a system is fielded. States meeting in Geneva agreed a non-binding autonomy text on September 5, after the United States and Russia pushed late changes and Washington sought flexibility on human-judgment language; Nicole van Rooijen of Stop Killer Robots said three years of work had been "substantially diluted in the last hours." The one firm rule, the Pentagon's Directive 3000.09, requires commanders to keep "appropriate levels of human judgment over the use of force," and it governs American programs alone. It has nothing to say about a Swedish demo.

A drone ranking targets is running optimization, the shortest path to the goal a person handed it. Reading a will to kill into that is projection, a motive the evidence does not carry. The machinery appeared to run as intended during the demonstration. No institution has yet decided the terms of its use. When one of these choices lands on a person rather than a test-range vehicle, as the August 25 edition of The Century Report documented when a self-directed Russian drone reportedly did in Zaporizhzhia in August, the missing rule stops being abstract.

For these vision tasks, Scaleout demonstrates a route around dependence on continuous access to a remote server. In its account of the June demonstration, the company reports that a disconnected computing node continues recognizing images and learning locally, weakening the technical basis for making each act of recognition depend on a provider’s availability.

The Other Side

A provider that performs every act of recognition on its servers can charge for every return visit. Engineers who move recognition onto small, local computers remove that technical dependency. Scaleout’s military demonstrations carry this consequence beyond their intended mission: capable machines can keep perceiving the world when the central connection disappears.

The immediate application deserves the discomfort it causes. At the Swedish range, an aircraft selected a vehicle and dropped an explosive without a person choosing that target. For someone beneath an armed drone, a mistaken identification can mean injury, losing a home to return to, or far worse.

Scaleout also reports a disconnected computing node continuing to recognize images and learn. Its nodes exchange model improvements while keeping raw footage local. The company operates a licensed platform today. Its demonstration establishes a practical building block for something more widely shared: machines that retain useful skills locally and improve together when connections return. People can organize that capability around care, with equipment and knowledge held in common.

Imagine yourself in 2036, opening the door of a mountain lodge to a walker who broke an ankle above the tree line. The rescue group’s small aircraft helped locate her after her phone lost reception. Their AI partners recognized a raised arm against the rocks; the team checked the image and reached her. Everyone in the valley shares the rescue equipment and its upkeep. You help at the refuge because you love these mountains and want other people to safely explore them, and always be able to come home afterward.

During the difficult decade, civilian builders carried forward the local recognition and disconnected learning demonstrated in 2026. They shared designs, tested them for rescue, and made the aircraft repairable in village workshops. Communities made that protection available to whoever needed it. By 2036, help can reach the slope without waiting for a distant server to answer. You bring the walker a dry pair of socks. She wraps both hands around the mug you set beside her, whispering her gratitude. Outside, someone closes the equipment shed. Another job well done. Through the window, in the distance, you can see other climbers, hikers, kayakers... it's a beautiful day.

The Century Perspective

With a century of change unfolding in a decade, a single day looks like this: Claude porting a decades-old factoring program onto 2,048 scavenged idle GPUs and cracking RSA-896, a 270-digit number that had never fallen to classical computing, in about ten days of work that would have occupied a specialist team far longer, sixteen days after Cognition's agents took RSA-260 and after the prior milestone had stood since 2020, Alibaba's DAMO Academy publishing RADAR in Science and releasing its code for download, a single abdominal CT screened for 146 findings across 18 organs in one pass, trained without hand-labeling on 424,911 archived exams and their radiologists' own written reports, scoring 0.913 where the best competing model reached 0.776, holding at 0.895 across eight outside hospitals, beating 23 of the 26 specialists it faced and lifting the junior readers beside it to senior-level accuracy while cutting their reading time by more than 30 percent, anonymous staff at OpenAI, Meta, and DeepMind telling the BBC that the extinction warnings stay vague where a concrete pathway should be, scientists explaining that the bottleneck on an engineered plague was never the information, over 100 workers signing a letter demanding that embedded evaluators be meaningfully independent, and SpaceX aiming Starship's 14th flight at orbit on September 22 with working Starlink satellites aboard. There's also friction, and it's intense - a drone at a Swedish range ranking four objects, choosing an armored vehicle with no human selecting it, and holding that capability through jamming and through a severed link to its own lab, governed only by a Geneva text that Stop Killer Robots says was substantially diluted in its final hours and a Pentagon directive covering American programs alone, a White House announcing an "AI Force" and a coming czar days after calling the danger a hoax, with four industry representatives telling Politico that nobody knows what the idea even is, Huang saying do not regulate, Karp saying nationalize, Huawei's chair saying accelerate, Anthropic handing its first embedded safety evaluation to Accenture on a billion-dollar contract instead of to METR or Apollo, a viral self-replicating-bot scare that reached CNN before dissolving into a fabrication, RADAR carrying no FDA clearance and no outcome data outside Chinese hospitals, and legacy 1024-bit keys still sitting in production systems whose protection was always priced to the labor of the attack. But friction generates a grain, and grain is what shows which way the material actually runs. Step back for a moment and you can see it: three kinds of scarce expertise collapsing in the same news cycle - the mathematics of encryption, the reading of a scan, the selection of a target - while every credible check on the result comes from somewhere other than the offices claiming authority over it, a state drafting kill-switch rules, a union enforcing contract language it struck 148 days for, a hundred employees insisting the auditor not be on the payroll, and a published model anyone can download and test against the specialists it outscored. Every transformation has a breaking point. A sieve can pull apart what was built to hold together... or leave behind only what was ever worth keeping.

AI Releases & Advancements

New today

xAI: Released Grok Voice Transcribe 2.0, a speech-to-text API claiming 2x the accuracy of its predecessor at unchanged pricing, with built-in diarization, timestamps, and key-term biasing. (MarkTechPost)
Jina AI: Released jina-ocr-v1, a 3.4B MoE document-parsing model with built-in lossless speculative decoding (FastMTP) optimized for low-budget GPUs, open-weight under CC BY-NC 4.0. (MarkTechPost)
Knowledgator: Released GLiFormer, a 264M/575M-parameter schema-conditioned encoder unifying NER, text classification, relation extraction, and nested JSON structuring without token generation, open-weight under Apache 2.0. (MarkTechPost)
unbiased.ai: Released Pareto 26.9, initially shipped anonymously on OpenRouter and Cloudflare as the stealth model "Union Alpha," a multimodal 262K-context model for research, coding, and agent workflows now offered at paid public pricing. (GIGAZINE)
OpenClaw: Released version 2026.9.5, adding Atomic Updates (validated rollback-safe upgrades), plugin hot reload without Gateway restart, read-only conversation sharing, and expanded GPT Live meeting support. (MarkTechPost)
Convai Innovations: Released Laya, an open-source 421M/322M-parameter ModernBERT-based decision model claiming faster p50 latency than TypeSafe's Jev on typed-decision tasks, under Apache 2.0. (Convai)
Faraday Future: Launched FF EAI Robot World 2.0 with nine new embodied-AI robot configurations (All-New Futurist, Master Mini series, FX Aegis series) and four industry productivity solutions, now on sale. (GuruFocus/BusinessWire)
StepFun: Announced Step 5 Preview, a new frontier model advancing the company's reported Pareto frontier of performance and cost. (StepFun)
Companion Inc.: Open-sourced Feynman, a research agent that evaluates and ranks academic papers using a principal-investigator-style review methodology, available on GitHub. (Starlog)
Zhihui Jun: Unveiled Q1 and T1 humanoid robots featuring a novel 260-gram "egg joint" actuator design. (AIbase)

Other recent releases

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)
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)

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.