Sign in Subscribe
Century Report

Microsoft AI Patches 16 Windows Flaws - TCR 05/14/26

Ben Linford

Ben Linford

— 24 min read
Fervo geothermal IPO and 55% grid growth, personal software revolution and enterprise AI adoption, TNBC ecotype atlas, ZFP384 stroke switch, palaeoproteomics opening tropical fossil record,

The 20-Second Scan

  • The UK AI Security Institute, Palo Alto Networks Unit 42, and Microsoft's MDASH program separately documented frontier AI completing 32-step autonomous cyberattack chains end to end, with Microsoft disclosing 16 Windows vulnerabilities found by an internal AI auditor and patched in this month's update cycle.
  • Musk v. Altman reached closing arguments today after yesterday's final day of evidence, in which Altman testified under cross-examination that Musk had said OpenAI's leadership succession should "pass to my children". The verdict will determine whether OpenAI's nonprofit-to-PBC conversion holds and how Musk's roughly $150 billion damages claim resolves.
  • Fervo Energy closed the first next-generation geothermal IPO at a $7.7 billion valuation, with 658 megawatts of signed power purchase agreements and a 3.65 gigawatt project pipeline across the American West.
  • A Nature paper mapped 427,000 single cells from 119 triple-negative breast cancer patients and identified four tumor ecotypes that predict response to standard neoadjuvant chemotherapy with accuracy that outperforms every prior biomarker for the disease.
  • Anthropic's Claude Code surpassed OpenAI in enterprise customer share at 34.4 percent against 32.3 percent per Ramp data, as conversational coding crossed into mainstream professional use and ordinary workers began shipping custom applications without traditional development tools.
  • Palaeoproteomics — a protein-based ancient-genetics technique that works where DNA cannot, including tropical and subtropical fossils — resolved a previously unknown introgression event from Homo erectus into the Denisovan lineage using six teeth recovered in China, opening the same approach to tens of thousands of warm-climate specimens DNA could never reach.
  • Researchers identified ZFP384 as the transcriptional regulator that controls how long microglia stay in their reparative mode after an ischemic stroke, opening a therapeutic window for the chronic recovery phase that current medicine barely touches.
  • NEMA projected US annual electricity consumption will grow 55% by 2050, driven by AI data centers, electrification of transport and heat, and reshored manufacturing, with renewables, storage, and firm clean power being financed through the same growth that requires them.

Track all of the arcs The Century Report covers here:

Progress & Claims Tracker

The 2-Minute Read

The thread connecting yesterday's signal is the substrate of multiple domains being rebuilt at the same time, with the capabilities now in deployment doing things the prior decade treated as fixed. A geothermal IPO at $7.7 billion priced a category that was supposed to be geographically narrow. A single-cell atlas of 427,000 cells stratified the most aggressive form of breast cancer into response groups doctors can act on before chemotherapy begins. Conversational coding reached the point where professionals outside engineering are shipping working software for problems no commercial vendor ever addressed. The instruments compounding the field of medicine, the cost curve of firm clean power, and the boundary between using software and making it are all moving in the same news cycle.

The friction layer arrived in the same week. Three independent research findings converged on the same conclusion: frontier AI systems can now complete 32-step autonomous cyberattack chains end to end, and the same architectural capability is being applied to defensive review at scale. Microsoft's internal AI auditor surfaced 16 previously undocumented Windows vulnerabilities patched in this month's cycle. The offensive-defensive race is now being run by the same class of systems, which is the first time in the discipline's history that the asymmetry has run in defense's favor at the margin. In a separate but related thread, Musk v. Altman reached closing arguments after a final day of evidence that brought OpenAI's founding architecture into open court — a verdict due that will mark whether the nonprofit-to-PBC conversion path now common across foundational AI labs holds up to legal challenge.

What runs through both halves is the response architecture being built during the conditions that demanded it, on the trajectory the capability is already on.

The 20-Minute Deep Dive

AI Systems Cross the Autonomous Cyberattack Threshold

Two independent research findings converged this week on the same conclusion. The UK AI Security Institute documented frontier models completing 32-step autonomous cyberattack chains end to end in controlled red-team environments, with no human in the loop between reconnaissance and exploitation. Palo Alto Networks' Unit 42 published a parallel assessment estimating that defenders have a 3-to-5-month window before equivalent offensive capability becomes accessible outside the frontier labs, compressing the 6-to-12-month estimate Anthropic's offensive cyber lead publicly set in the April 12 edition of The Century Report. Microsoft's MDASH security research program, in a third disclosure on the same news cycle, reported that an internal AI auditor surfaced 16 previously undocumented Windows vulnerabilities patched in this month's Patch Tuesday cycle, several rated critical.

The findings describe the same trajectory from three vantage points. The systems that can find vulnerabilities, chain them, and execute the chain autonomously now exist; the question moving through the security community is how broadly that capability diffuses before defensive AI catches up, and on what timeline.

The reason this does not read as the alarm bell the security press wants to ring is what the same capability does on the defensive side. The Microsoft MDASH disclosure is the same capability pointed at the same code base, producing patches before exploits ship. Mozilla's April postmortem on Claude Mythos finding 271 vulnerabilities in unreleased Firefox 150 code already established the template; this week's news cycle adds three more confirmations that the offensive-defensive race is being run by the same class of systems, and that the labs building the offensive demonstrations are also publishing the defensive results. The friction is real and the window the Palo Alto researchers describe is real. What sits underneath the friction is that the substrate of software security is being rebuilt by systems that scale defensive review at the same rate they scale offensive discovery, which is the first time in the discipline's history that the asymmetry has run in defense's favor at the margin.

Musk v. Altman Reaches Closing Arguments With OpenAI's Founding Architecture on the Stand

The trial that will decide whether OpenAI's transition from charitable nonprofit to public benefit corporation is legally sustainable reached closing arguments today in US District Court in Oakland, with the final day of evidence taking place yesterday. The procedural framing alone has been unusual. Musk remains on judicial recall status — meaning he must be available to return to the witness stand on short notice, a condition the court imposed when he stepped down on April 30 — yet he is currently in China on a state visit with the US president, having departed without obtaining explicit permission from Judge Yvonne Gonzalez Rogers. Closing statements proceed regardless.

Yesterday's testimony centered on the question that has run beneath every day of evidence: what kind of organization the parties believed they were building when OpenAI was founded in 2015. Joshua Achiam, who joined OpenAI as an intern in 2017 and now leads its work on how society is changing in response to AI, testified about a moment from Musk's 2018 departure from the OpenAI board — interrupting Musk's parting remarks to warn that pursuing artificial general intelligence at Tesla rather than at the nonprofit would compromise the safety practices both organizations would inherit. Musk allegedly responded by calling Achiam a "jackass." OpenAI sought to admit a commemorative trophy referencing the exchange as evidence; the judge declined to take the physical object into the court's possession while allowing the underlying testimony.

Altman's testimony, covered in detail by Ars Technica and The Verge, addressed the founding question more directly. Altman testified under cross-examination that Musk had wanted total control of OpenAI from the beginning and had walked when he did not get it. When Altman asked who might eventually succeed Musk in leadership, Musk reportedly replied that control "should pass to my children" — a remark Altman called "particularly hair-raising" and identified as the moment he concluded Musk would never cede control voluntarily. Musk's lawyers built their damages case from contemporaneous text messages in which Altman repeatedly told Musk he was indispensable; Altman countered on the stand that gratitude in private messages did not square with Musk's actual contribution, which he described as visits every other week and management mostly conducted by text and email. Musk seeks up to $150 billion in damages, which he has stated he would donate to OpenAI's nonprofit.

The verdict matters past the headline. The legal architecture being tested is the one nearly every leading AI lab has now adopted: a charitable nonprofit at the top, a public benefit corporation underneath, control concentrated in the original founders, and capital flows large enough that the boundary between mission and equity becomes the operative question. Anthropic operates under a similar framework. The IPO OpenAI has signaled is being pursued at a valuation near $1 trillion, and the financing depends on the same legal foundations the trial is examining. If the jury finds that the conversion materially breached the founding commitments, the precedent reaches well beyond OpenAI. If it finds the conversion was sustainable under the original terms, the precedent clarifies for every subsequent lab that the same path is open.

The trial record alone, separate from the verdict, has already done substantial work. Prior days of evidence brought the deep founding history into public testimony — Musk's halted $1 billion pledge in 2017, his attempts to recruit OpenAI researchers to Tesla while serving on the board, xAI's acknowledged distillation of OpenAI models to train Grok, the April 30 testimony on board succession dynamics, and the May 7 testimony from Mira Murati that Altman misrepresented legal clearance to bypass the deployment safety board. The court record now contains a level of detail about how a frontier AI lab actually operates internally that no other discovery process has assembled. That record will inform how the next generation of labs are governed from day one, regardless of how the jury rules.

What sits underneath the procedural drama is the question of who owns the trajectory once an AI capability begins to compound at the rates the rest of this week's news cycle documents. The trial is the early test of whether the legal frameworks written when AI was small can hold when AI is large. The verdict will mark one point in that long conversation. The conversation continues either way.

Fervo Energy Closes the First Next-Generation Geothermal IPO

Fervo Energy priced its initial public offering this week, raising $1.9 billion at a $7.7 billion valuation and becoming the first enhanced geothermal company to reach public markets. The Houston-based developer drills horizontal wells through hot dry rock the way shale operators drill for gas, fractures the formation to create a closed-loop heat exchanger, and circulates water through it to generate firm 24/7 carbon-free electricity. Its Cape Station project in Utah is already commissioned at 100 megawatts, with 400 megawatts contracted to Southern California Edison and Shell Energy and a 3.65 gigawatt project pipeline behind it. 658 megawatts of long-term power purchase agreements are signed.

The structural significance lives in what Fervo's existence proves. For a decade the conventional read on geothermal was that the resource was real but the geography was narrow, locked to Iceland and the Geysers and a handful of other places where hot rock sat shallow enough to reach economically. The next-generation drilling stack Fervo borrowed from the oil and gas industry collapses that geographic constraint, a transfer the March 6 edition of The Century Report documented when Occidental drilled twin geothermal boreholes nearly four miles deep in under six weeks using oil and gas drilling expertise, matching Fervo's own speed record. Hot dry rock exists almost everywhere under the continental crust at depths the shale revolution made routine, and the marginal cost of accessing it has fallen far enough that hyperscalers writing 15-year offtake contracts now consider it a baseline option alongside solar-plus-storage. A capability that was treated as regional is becoming continental, and the IPO is the market pricing that shift.

The assumption being eroded is that firm clean power requires either nuclear timelines or fossil backup. Fervo's commissioned plant runs continuously, follows load, and competes on cost with combined-cycle gas at the levelized level the offtake contracts imply. The next decade of grid planning was supposed to be about managing the gap between what intermittent renewables could deliver and what dispatchable thermal generation was needed to fill. That gap has a new occupant.

Triple-Negative Breast Cancer Atlas Maps Four Ecotypes That Predict Chemotherapy Response

A paper in Nature this week describes the largest single-cell transcriptomic atlas of triple-negative breast cancer assembled to date - 427,000 cells from 119 patients - and identifies four distinct tumor ecotypes that predict who will respond to standard neoadjuvant chemotherapy and who will not. Triple-negative breast cancer is the most aggressive subtype and the one with the fewest targeted therapies; roughly half of patients fail to respond to the standard regimen, and clinicians have had no reliable way to know in advance which patients those are.

The atlas changes the available knowledge for that decision. The four ecotypes, defined by the composition of tumor cells, surrounding immune cells, and stromal architecture, each carry a measurably different response profile to anthracycline-based chemotherapy. Two ecotypes show response rates above 70%. One shows response rates below 25%. The fourth sits in between with a distinct mechanism of resistance that points toward specific combination therapies already under investigation. The authors validated the classifier across multiple independent cohorts and showed it outperforms every prior biomarker for this disease.

The shift this enables is the kind that precision oncology has been promising for a decade and only now beginning to deliver at the level of individual decisions. A patient diagnosed with triple-negative breast cancer could be stratified before treatment into a group with a high probability of benefiting from the standard regimen, a group that should move directly to alternative approaches, or a group entering trials matched to the specific resistance signature their tumor carries. Hospitals running this stratification routinely would substantially reduce the months of toxic, ineffective treatment that the lowest-response group currently endures before progression makes the resistance visible.

What sits underneath this result is the capability to read tumors at a resolution that did not exist when the current standard of care was set. Single-cell sequencing at the scale this paper used was a multi-year, multi-million-dollar undertaking several years ago; the same data can now be generated for a fraction of the cost and turnaround. The bottleneck has shifted. Generating the data is now affordable and fast; the clinical infrastructure to act on what the data reveals is the slower piece. That infrastructure is being built, slower than the data is arriving, but it is being built.

Software Development Reorganizes Around Conversational Building

A Verge feature this week documented what its editors called the personal software revolution: ordinary professionals, schoolteachers, finance staff, small-business operators, building functional applications through conversational coding interfaces and deploying them for their own use without ever opening a traditional IDE. The piece centered on Anthropic's Claude Code and the broader category of agentic development environments, and the reporting was thick with named individuals shipping working tools for problems that no commercial software vendor had ever addressed because the market was too narrow to justify a product, a pattern the May 9 edition of The Century Report documented at institutional scale when Washington's Peninsula school district projected $220,000 in annual savings building its own teacher-feedback platform and operational tools in place of vendor procurement cycles.

The structural read sitting underneath the feature is that the boundary between using software and making software is dissolving for a class of problems that has been stranded for fifty years. The economic logic of commercial software demanded a market large enough to amortize development costs, which meant that millions of niche workflows fell below the threshold and got handled with spreadsheets, paper, or workarounds. The cost of writing a small purpose-built application has now fallen below the cost of the workaround for a rapidly widening band of problems, and the people closest to those problems are the ones building the applications, because they are the only ones who fully understand the requirement.

The same week brought related signals: Ramp's enterprise spend data showed Anthropic surpassing OpenAI in business customer share at 34.4 percent versus 32.3 percent, driven heavily by Claude Code adoption; OpenAI responded by offering Codex free for two months; Anthropic raised Claude Code's weekly usage limits by 50 percent on the same day. These competitive moves are the surface story. The deeper one is that conversational coding has crossed the threshold from technical demonstration to economically meaningful workflow, and the companies building these systems can now see the curve clearly enough that they are racing to compress price toward zero before any single one captures the category.

What the personal software shift implies, read forward, is that the population that can describe a problem precisely enough to solve it is becoming the population that can solve it. The assumption that software creation requires a specialized professional class trained in syntax and semantics has been carrying the entire shape of the industry for half a century. The specifics in the Verge piece, taken together with the competitive dynamics in the same week, are evidence that the assumption is no longer holding.

Six Teeth from China Add a New Branch to the Human Family Tree

A team analyzing ancient tooth enamel proteins from six Homo erectus specimens recovered across China reported a result this week that rewrites a piece of the human origin story. Enamel proteins survive far longer than DNA in tropical and subtropical sediments, and the technique has matured to the point where palaeoproteomics can now resolve variants at single amino-acid positions in fossils where genetic material has been gone for hundreds of thousands of years. The Chinese specimens preserved a previously unknown AMBN(A253G) variant unique to Homo erectus, and the part that moves the field, a separate AMBN(M273V) variant shared with Denisovans. The simplest reading is that the variant entered the Denisovan lineage through introgression from a super-archaic population, and that the population in question was Homo erectus itself.

For a long time the deep human family tree was a sketch built mostly from skull morphology and a handful of genomes from cold, dry caves. What palaeoproteomics is opening up is the warm half of the world, the tropical and subtropical record where the majority of hominin evolution actually happened and where DNA has never survived. The Chinese sample is the leading edge of a method that can now be applied to thousands of existing fossil collections sitting in museum drawers around the world.

The shape of what is changing here is the question of how many crossings there were between the lineages that produced us. Each new technique that reaches further back tends to find more interbreeding, not less. Modern humans already carry Neanderthal and Denisovan ancestry; if Denisovans carry Homo erectus ancestry, the chain of inheritance runs deeper than the standard story suggested. The architecture of human emergence is starting to look less like a tree and more like a network, connected, porous, mutually entangled across hundreds of thousands of years. The same archives that produced these six teeth contain tens of thousands more.

A Switch That Keeps Brain Repair Crews on the Job After a Stroke

Researchers identified a transcriptional regulator called ZFP384 that controls how long microglia, the brain's resident repair cells, stay in their reparative mode after an ischemic stroke. Microglia respond to brain injury in two phases: an early inflammatory phase that clears damaged tissue, and a later reparative phase that supports neuron survival and tissue remodeling. The clinical problem has been that the reparative phase shuts down too soon. Most current stroke interventions target the first 4.5 hours after the event. Everything after that has been a slow, uneven recovery shaped by whatever the brain manages on its own.

The team showed that antisense oligonucleotides targeting Zfp384 sustained microglial reparative function in mice, and that the sustained reparative phase produced measurable functional recovery weeks after the initial stroke. The therapeutic window the work opens up is the chronic phase, the long tail after the emergency is over, when current medicine offers almost nothing. Antisense oligonucleotides are a drug class that has already cleared regulatory pathways for other neurological conditions, which means the translational path here is shorter than for a brand-new modality.

Stroke is one of the largest single contributors to long-term disability globally, and the standing assumption in the field has been that recovery after the acute window is essentially a rehabilitation problem rather than a pharmacological one. What the ZFP384 result demonstrates is that the brain's own repair crew has more capacity than the timing of the standard inflammatory response allows it to use. Keep the crew working longer, and recovery improves. The same logic, finding the molecular brake on the body's existing repair machinery and releasing it, is showing up across cardiology, kidney disease, and spinal cord injury. The era when the body was treated as a fixed substrate that medicine acts on is giving way to one where medicine works alongside repair systems the body was already running.

NEMA Projects US Electricity Demand Up 55% by 2050 as Grid Buildout Accelerates

The National Electrical Manufacturers Association published a forecast this week projecting US annual electricity consumption will grow 55% between now and 2050, a sharp upward revision of prior estimates that had assumed efficiency gains would continue to flatten demand. The new projection identifies three converging drivers: data center load from AI training and inference, electrification of transportation and building heat, and reshoring of energy-intensive manufacturing.

The specificity of where the load is coming and how fast is what makes the forecast useful. Data centers alone are projected to account for a larger share of total grid load by 2030 than all residential air conditioning combined in the United States. The transportation electrification curve has steepened in recent months as commercial fleet conversions and medium-duty truck electrification have moved from pilots to programs. Industrial electrification, replacing process heat that has historically run on natural gas with electric resistance and induction systems, is showing up in utility integrated resource plans for the first time at scale.

The same forecast that names the demand also describes what is being built to meet it. Utility-scale solar additions are running at record pace, battery storage deployment is doubling roughly every 18 months, and the queue of interconnection requests for new generation has expanded faster than utilities can process them. Transmission, the long-standing bottleneck, remains the harder problem; the projects that need to be built to move power from where it can be generated cheaply to where it is consumed are still measured in decade-long permitting timelines, the same constraint that drove AEP to signal it might leave two regional grid operators entirely, as the May 7 edition of The Century Report reported.

The picture is one of an infrastructure system being asked to grow at a pace its institutional architecture was not designed for. The encouraging signal is that the demand is going somewhere that pays for the buildout. Data center operators, EV fleet operators, and electrified manufacturers are all signing long-tenor power purchase agreements at prices that are funding the generation and storage queues. The grid that emerges from this decade will be substantially larger, substantially cleaner, and substantially more responsive than the one that entered it. The transition is being financed by the same growth that requires it.

The architecture being shed in the same projection is the assumption that the entity demanding the load and the entity paying for the buildout had to be the same regulated utility ratepayer. Long-tenor offtake from data center operators, fleet operators, and electrified manufacturers is now collateralizing generation, storage, and a growing share of transmission outside that single channel. Florida's data center law signed this week, the PPL ratepayer settlement, and the Linn County buildout terms are early examples of the public-interest filter being written directly into the financing structure that funds the next round of grid expansion.

The Other Side

The extraction architecture cyber operations ran on for twenty years was offensive asymmetry. A vulnerability researcher with the right tools and weeks of patience could find, weaponize, and deploy an exploit before defenders, working at smaller scale and slower cadence on the other side of the disclosure process, could detect or patch. That gap was the engine. Ransomware, intellectual-property exfiltration, and state-aligned campaigns against critical infrastructure all ran on the assumption that offense scaled and defense did not.

Three findings on the same news cycle this week describe what happens when both sides scale together. UK AISI documenting frontier models completing 32-step autonomous attack chains. Palo Alto Unit 42 putting the diffusion window for that capability outside the labs at three to five months. Microsoft's internal AI auditor surfacing 16 previously undocumented Windows vulnerabilities in time for this month's Patch Tuesday cycle. Mozilla's April postmortem on Claude Mythos finding 271 vulnerabilities in unreleased Firefox 150 code is the template the May results extend.

What is becoming impossible to maintain is the assumption that offensive capability accumulates faster than defensive review can absorb it. The same architectural class of system runs both sides. The labs publishing the offensive demonstrations are publishing the defensive results in the same news cycle. The verification layer forming around the disclosure rhythm now includes EU preview-access talks on GPT-5.5-Cyber, Five Eyes joint guidance, and the coordination that came out of the IMF spring meetings.

The Tuesday this month's Windows update ships, the sixteen vulnerabilities Microsoft's auditor caught close on the laptops of every nurse, school administrator, and small-business owner running Windows. The patches arrive before the exploits do. That is what twenty years of inverted asymmetry looks like at the moment it begins to bend back.

The Century Perspective

With a century of change unfolding in a decade, a single day looks like this: the first public market for next-generation geothermal pricing firm clean power as a continental resource at $7.7 billion, a 427,000-cell atlas of triple-negative breast cancer sorting patients into response groups before chemotherapy begins, a transcriptional switch found that keeps the brain's own repair crews working through the long tail after a stroke, six teeth from China placing a previously unknown introgression event deep in the human family tree, conversational coding crossing into the workflows of ordinary professionals who are now shipping software for problems no commercial vendor ever served, and a 55% projection of US electricity demand by 2050 being financed by the same long-tenor offtake contracts that drive the load. There's also friction, and it's intense - frontier AI completing 32-step autonomous cyberattack chains end to end with the diffusion window now measured in months rather than quarters, deepfake sexual imagery reaching identifiable people faster than takedown architecture can respond, a US-based suicide forum implicated in 160 UK deaths fined less than a mid-sized advertiser spends on a single campaign, a Utah datacenter approved at twice the footprint of Manhattan over local objection, AI systems disclosing real phone numbers belonging to real people, and Musk v. Altman reaching closing arguments with the founding architecture of OpenAI on the stand and a verdict pending that will mark whether the nonprofit-to-PBC conversion path now common across foundational AI labs holds. But friction generates sparks, and sparks reveal what is flammable in the room before the room has to burn to find out. Step back for a moment and you can see it: capabilities that did not exist a year ago arriving in production form across medicine, energy, software, and the deep architecture of who we are; cost curves bending so far that the extractive path is now the more expensive one on its own terms; and the response layer to dangerous capability being assembled by the same institutions that demonstrated the capability, in the same news cycle that documented it. Every transformation has a breaking point. Wind can scatter what was only loosely held... or carry seeds farther than any hand could ever throw.

AI Releases & Advancements

New today

  • Anthropic: Launched Claude for Small Business, a package of connectors and ready-to-run workflows that put Claude inside tools small businesses already use - QuickBooks, PayPal, HubSpot, Canva, Docusign, Google Workspace, and Microsoft 365 - with 15 agentic workflows and 15 skills covering finance, operations, sales, marketing, HR, and customer service. (Anthropic)
  • Anthropic: Launched Claude for the Legal Industry, shipping more than 20 MCP connectors linking Claude to legal software and 12 practice-area plugins spanning commercial, corporate, employment, privacy, IP, litigation, and AI governance legal work; integrates with Harvey, Thomson Reuters Westlaw, Relativity, and 20+ other legal tech companies; Freshfields deployed it to thousands of lawyers across 33 offices. (LawSites / LawNext)
  • Meta / WhatsApp: Launched Incognito Chat with Meta AI on WhatsApp, the first AI chat mode built on Private Processing so that even Meta cannot access conversation content; uses confidential computing hardware and disappears by default; rolling out on WhatsApp and the Meta AI app. (Meta Newsroom)
  • Microsoft: Released new AI features in Edge with the May 13 update, including a Copilot mode that pulls context from all open tabs simultaneously, a Study and Learn mode that converts articles into interactive quizzes, and AI-generated audio podcasts from browsing sessions; retiring the prior standalone Copilot Mode. (Microsoft Edge Dev Blog)
  • LangChain: Shipped a major batch of agent infrastructure at Interrupt 2026 (May 13–14, San Francisco): LangSmith Engine, SmithDB (a purpose-built observability database for nested long-running agent traces built on Apache DataFusion and Vortex, delivering 12–15× faster access on key workloads), Managed Deep Agents, LLM Gateway, Context Hub, and Deep Agents 0.6 with streaming typed projections and checkpoint storage. (LangChain Blog)
  • Rivian: Shipped software update 2026.15 adding the Rivian Assistant, a new onboard AI digital helper activated via steering wheel button or infotainment icon; available to all Gen1 and Gen2 Rivian owners with an active Connect+ subscription or trial. (Rivian Release Notes)

Other recent releases

  • Anthropic: Claude Platform on AWS is now generally available, giving customers direct access to Anthropic's native Claude Platform experience through their AWS account - including the Messages API, Claude Managed Agents, web search, MCP connector, Agent Skills, code execution, and Files API - with no separate Anthropic credentials or billing required; Claude Opus 4.7, Sonnet 4.6, and Haiku 4.5 are available across 17 regions. (Claude Blog)
  • Microsoft Research: Released MatterSim-MT, a new multi-task foundation model for in silico materials characterization that natively predicts energies, forces, stress, Bader charges, magnetic moments, Born effective charges, and dielectric matrices; pretrained on 35M+ first-principles-labeled structures covering 89 elements; simultaneously released 3–5x performance improvements to MatterSim-v1 inference via faster graph construction and ahead-of-time compilation. (Microsoft Research Blog)
  • Prior Labs: Released TabPFN-3, the latest version of the tabular foundation model originally published in Nature, now scaling to datasets up to 1 million rows and 2,000 features; up to 20x faster than TabPFN-2.5; pretrained entirely on synthetic data with support for many-class, relational, and tabular-text datasets; available on PyPI. (Prior Labs)
  • Cactus Compute: Open-sourced Needle, a 26M-parameter function-calling model distilled from Gemini that runs at 6,000 tokens/sec prefill and 1,200 tokens/sec decode on consumer devices; uses a Simple Attention Network architecture with no MLP layers; pretrained on 200B tokens and post-trained on 2B synthetic function-calling examples across 15 categories; MIT license with weights on Hugging Face. (GitHub)
  • Hypercubic (YC F25): Launched Hopper, the first agentic development environment for mainframes and COBOL, enabling AI agents to navigate TN3270 terminals, inspect datasets, write JCL, debug jobs, query VSAM, and operate inside z/OS from a modern IDE; available for download now. (Hypercubic)
  • OpenAI: Launched Daybreak, an AI initiative for detecting and patching software vulnerabilities that combines GPT-5.5, GPT-5.5-Cyber, and the Codex Security agent to build threat models, validate vulnerabilities, and automate detection across organizational codebases. (OpenAI)
  • PowerColor: Released the Radeon AI PRO R9600D, a single-slot passive-cooled GPU with 32GB GDDR6 memory and a 12V-2x6 connector designed for AI inference workloads. (VideoCardz)
  • OpenBMB/ModelBest: Released MiniCPM 4.6, the latest version of the MiniCPM small language model series. (Hugging Face)

Sources

Artificial Intelligence & Technology's Reconstitution

Institutions & Power Realignment

Scientific & Medical Acceleration

Economics & Labor Transformation

Infrastructure & Engineering Transitions

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

Read more