Sign in Subscribe
Century Report

Mythos Enters Government - TCR 04/17/26

Ben Linford

Ben Linford

— 16 min read
Four-panel infographic: White House distributing Claude Mythos to federal agencies, Opus 4.7 staged release architecture, robot and biomedical breakthroughs, grid battery and solar scaling.

The 20-Second Scan

  • The White House is preparing to distribute Claude Mythos to major federal agencies including Defense, Treasury, Commerce, Homeland Security, Justice, and State, weeks after designating Anthropic a supply-chain risk.
  • Finance ministers, central bank governors, and bank CEOs at the IMF spring meetings in Washington described Mythos as warranting the coordinated attention of every finance ministry, with Anthropic confirming UK banks will receive access within days.
  • Anthropic released Claude Opus 4.7, its most capable generally available model, while explicitly noting it scores lower than Mythos Preview on every evaluation and is being used to test cybersecurity safeguards before broader Mythos-class deployment.
  • Physical Intelligence published research showing its π0.7 model directed robots to perform tasks they were never explicitly trained on, demonstrating compositional generalization from minimal data fragments.
  • University of Oklahoma researchers identified FGF21 as a hormone that reverses obesity in mice through a hindbrain circuit overlapping with GLP-1 drug targets but operating through a completely different mechanism - energy burning rather than appetite suppression.
  • UCLA scientists reversed fatty liver disease in mice by clearing senescent "zombie" macrophages that accumulate with age and cholesterol, even without dietary changes, with the same molecular signature confirmed in human liver biopsies.
  • New England's largest grid battery came online at 250 MW in Medway, Massachusetts, with a 700 MW successor already in development and state clean-peak policy driving a wave of multihundred-megawatt storage projects across the region.
  • US community solar crossed 10 GW of installed capacity with over 8 GW of additional projects in the pipeline, though installations fell 25% in 2025 as mature markets saturated and new state programs prepare to absorb the next wave.

Track all of the arcs The Century Report covers here:

Progress & Claims Tracker

The 2-Minute Read

The institutional architecture surrounding Mythos crossed a structural threshold yesterday. The same administration that designated Anthropic a supply-chain risk is now preparing to distribute the model it tried to ban to six Cabinet-level departments. The Canadian finance minister compared the challenge to the Strait of Hormuz - except "the issue that we're facing with Anthropic is that it's the unknown unknown." The Bank of England governor, the ECB president, and Barclays' CEO all spoke publicly about the model at the IMF meetings, while Anthropic confirmed UK banks will receive access within days. When a single intelligence system's capabilities become a standing agenda item at the most significant gathering of global financial officials, the governance architecture for frontier AI is being built under conditions that leave no room for deliberation at the pace institutions are accustomed to.

Anthropic's simultaneous release of Opus 4.7 - explicitly positioned as less capable than Mythos on every benchmark - reveals a new deployment architecture for dangerous capability. The company is using its generally available model as a testing ground for cybersecurity safeguards that will eventually gate broader Mythos-class access. This is staged release as governance: rather than choosing between withholding capability and distributing it widely, Anthropic is building a pipeline where each model generation serves as a proving ground for the safety infrastructure of the next. The approach carries real implications for how frontier capability reaches the world - gradually, through layers of tested constraint rather than through a single launch decision.

Physical Intelligence's π0.7 research and the twin metabolic discoveries from Oklahoma and UCLA occupy different domains but share a structural signature: systems revealing capabilities that exceed what their inputs would predict. A robot trained on two episodes involving an air fryer - neither of which demonstrated actual cooking - synthesized those fragments into functional understanding of the appliance. A hormone that targets the same brain region as GLP-1 drugs achieves weight reversal through the opposite mechanism. Senescent immune cells that accumulate with age and cholesterol drive fatty liver disease, and removing them reverses the damage even on an unchanged diet. In each case, the finding opens a pathway that existing frameworks did not anticipate. The distance between what we assumed was possible and what these systems demonstrate is where the transition's acceleration is most visible - not in headline announcements, but in results that surprise even the researchers who produced them.

The 20-Minute Deep Dive

Mythos Enters the Rooms Where Civilization Is Administered

The White House memo reported by Bloomberg yesterday represents one of the most striking institutional reversals The Century Report has tracked. Gregory Barbaccia, the federal chief information officer, notified officials at the Departments of Defense, Treasury, Commerce, Homeland Security, Justice, and State that the Office of Management and Budget was establishing protections to allow agencies to begin using Mythos in the coming weeks. The memo's subject line was "Mythos Model Access." This is the same administration that, six weeks ago, designated the company that built Mythos a supply-chain risk under a statute previously reserved for foreign-linked entities.

The reversal did not happen because the legal dispute was resolved. The D.C. Circuit's split ruling remains unresolved, with May 19 oral arguments still on the calendar. The reversal happened because the capability proved too consequential to exclude from institutional defense. As The Century Report documented across the past ten days, Goldman Sachs and JPMorgan both disclosed Mythos deployments on earnings calls, the U.S. Treasury convened bank CEOs, and UK regulators coordinated emergency assessments. The Bloomberg report confirms that Anthropic briefed senior government officials on both offensive and defensive applications before the corporate rollout - meaning the administration was aware of the capability's scope before it designated the company a threat.

The IMF spring meetings in Washington amplified this dynamic to a global stage. Canada's finance minister François-Philippe Champagne told the BBC that the situation "is serious enough to warrant the attention of all the finance ministers," distinguishing it from the Strait of Hormuz by noting that at least with physical geography, "we know where it is and we know how large it is." The ECB's Christine Lagarde described Anthropic as "a responsible company that is suddenly thinking: 'Ah, that could be really good' - but if it falls in the wrong hands, it could be really bad," before acknowledging that no governance framework exists to "actually mind those things." Bank of England Governor Andrew Bailey asked the central question for regulators everywhere: "What is the optimum moment to frame the rules of the road? If you go too early you risk missing the target and distorting the evolution, and if you go too late things can get out of control."

Barclays CEO CS Venkatakrishnan offered the most succinct framing: "This is what the new world is going to be." Anthropic's UK head Pip White confirmed that British banks would receive Mythos access "in the very near term, in the next week," adding that engagement from UK CEOs "has been significant." The Guardian reported that the UK's AI Security Institute remains one of the few organizations globally to have published an independent evaluation of the model, and Anthropic's London expansion to 800 staff - documented in yesterday's Century Report - now reads as infrastructure preparation for precisely this kind of institutional integration.

What makes this moment structurally distinct from prior technology governance challenges is the compression. From Anthropic's first public acknowledgment of Mythos on April 7 to the White House preparing federal distribution is ten days. From Project Glasswing's launch to coordinated discussions among finance ministers of the world's largest economies is under two weeks. The governance frameworks that took years to develop for nuclear technology, decades for financial derivatives, and over a decade for internet regulation are being compressed into days and weeks - and the institutions involved are openly acknowledging that they are operating without adequate frameworks while doing so. This extends the split between broad and tightly restricted cyber-model deployment that the April 16 edition of The Century Report traced through OpenAI's GPT-5.4-Cyber release and Anthropic's eleven-organization Mythos strategy.

Opus 4.7 and the Architecture of Staged Release

Anthropic's release of Claude Opus 4.7 carries more structural significance than its benchmark scores suggest. The model retakes the top position among publicly available systems for agentic coding, scoring 64.3% on SWE-bench Pro and SWE-bench Verified. It improves on Opus 4.6 for complex engineering tasks, image analysis, instruction-following, and document creation.

The revealing detail is what Anthropic chose to say about its relationship to Mythos. The system card states that Opus 4.7 does not advance the company's "capability frontier" because Mythos received higher results "on every relevant evaluation." Anthropic wrote: "We stated that we would keep Claude Mythos Preview's release limited and test new cyber safeguards on less capable models first. Opus 4.7 is the first such model." The company noted that during training, it "experimented with efforts to differentially reduce" Opus 4.7's cyber capabilities, and that the model scored 73.1% on vulnerability reproduction versus 73.8% for its predecessor - a deliberate step backward in one domain to test constraint mechanisms.

This represents a new pattern in how frontier capability reaches the world. Rather than a binary choice between release and restriction, Anthropic is building a pipeline where each generally available model serves as the proving ground for the safety infrastructure of the next, more capable system. Security professionals who want Opus 4.7's full cyber capabilities can join a new Cyber Verification Program that relaxes certain safeguards under verified conditions. The findings from that program will inform "our eventual goal of a broad release of Mythos-class models." It is a more formal version of the staggered defender-first architecture that April 8 of The Century Report documented when Project Glasswing gave critical infrastructure operators exclusive early access to Mythos.

Early customers including Intuit, Harvey, Replit, Cursor, Notion, Shopify, Vercel, and Databricks tested Opus 4.7 before release. Pricing remains unchanged from Opus 4.6 at $5 per million input tokens and $25 per million output tokens. The model is available across all Claude products and through the API. What Anthropic is selling, though, extends beyond any single model's capabilities. The company is demonstrating an institutional capacity to manage the release of dangerous capability through graduated constraint - and using that demonstration as a commercial differentiator at the exact moment when global financial institutions are deciding which AI providers to trust with their infrastructure.

A Robot Brain That Learns Beyond Its Data

Physical Intelligence's π0.7 research, published yesterday, describes something that its own creators say surprised them. The model directed robots to perform tasks they were never explicitly trained on - a capability the field calls compositional generalization, where skills learned in separate contexts combine to solve novel problems.

The most striking demonstration involved an air fryer. When the research team investigated the training data, they found exactly two relevant episodes in the entire dataset: one where a different robot merely pushed an air fryer closed, and one from an open-source dataset where yet another robot placed a plastic bottle inside one. The model had somehow synthesized those fragments, plus broader web-based pretraining data, into a functional understanding of how the appliance works. With zero coaching, it made a passable attempt at cooking a sweet potato. With step-by-step verbal instructions - a human walking the robot through the task the way you might explain something to a new employee - it performed successfully.

Sergey Levine, co-founder and UC Berkeley professor, drew a direct comparison to the emergence of unexpected capability in language models: "Once it crosses that threshold where it goes from only doing exactly the stuff that you collect the data for to actually remixing things in new ways, the capabilities are going up more than linearly with the amount of data." Researcher Ashwin Balakrishna described buying a random gear set and asking the robot to rotate it: "It just worked."

The team measured π0.7 against its own previous specialist models - purpose-built systems trained on individual tasks - and found the generalist model matched their performance across complex work including making coffee, folding laundry, and assembling boxes. The implication, if it holds up to external validation, is that robotic AI may be approaching the kind of inflection point where capabilities begin compounding in ways that exceed what the underlying training data would predict. This follows the physical-AI reliability threshold that the April 7 edition of The Century Report tracked through Generalist's GEN-1 system reaching 99% success rates on delicate manufacturing tasks. This is the pattern that language models demonstrated several years ago. Seeing it in physical manipulation, where the gap between digital training and real-world execution has always been widest, represents a structural shift in what kinds of intelligence can generalize beyond their inputs.

Two Metabolic Pathways That Upend Assumptions

Two independent findings published yesterday identified mechanisms that challenge prevailing models of how the body manages weight and aging - and both open therapeutic pathways that existing frameworks had not anticipated.

At the University of Oklahoma, researchers found that FGF21, a naturally occurring hormone already attracting clinical attention, reverses obesity in mice by signaling through the hindbrain - the same region targeted by GLP-1 drugs like semaglutide. The surprise is that FGF21 operates through a completely different mechanism. Where GLP-1 medications reduce appetite and food intake, FGF21 increases metabolic activity, helping the body burn more energy. The hormone interacts with the nucleus of the solitary tract and the area postrema, which then communicate with the parabrachial nucleus. Lead researcher Matthew Potthoff noted: "We thought we would find that it signaled to the hypothalamus, so we were very surprised to discover that the signal was to the hindbrain."

FGF21 analogues are already in clinical trials for MASH, a serious form of fatty liver disease. Identifying the specific brain circuit through which FGF21 operates opens the possibility of more targeted therapies that avoid the gastrointestinal side effects and bone loss that current FGF21 analogues can produce. This extends The Century Report's tracking of the GLP-1 revolution - which has documented oral formulations, genetic resistance variants, psychiatric effects, and now an entirely parallel metabolic pathway operating through the same brain architecture via different biochemistry. It also extends the anti-obesity discovery compression that the April 13 edition of The Century Report tracked when Stanford's Peptide Predictor surfaced BRP as a non-GLP-1 appetite-regulating peptide and identified a genetic GLP-1 resistance mechanism in roughly 10% of the population.

At UCLA, researchers made a discovery about cellular senescence that may reframe fatty liver disease as a problem of immune system aging rather than diet alone. The team identified a molecular signature - the combination of two proteins, p21 and TREM2 - that reliably marks macrophages that have become senescent: still alive, still active, but no longer functioning properly and flooding tissues with inflammatory signals. In young mice, only about 5% of liver macrophages carried this signature. In older mice, the figure rose to 60-80%.

When the team treated mice with ABT-263, a drug that selectively eliminates senescent cells, liver size dropped from about 7% of body weight to a healthier 4-5%, and body weight fell roughly 25% - even though the animals continued eating a high-fat, high-cholesterol diet. The same senescent macrophage signature was significantly elevated in diseased human liver biopsies compared to healthy ones. The finding that excess cholesterol itself can push macrophages into senescence suggests a mechanism by which unhealthy diets accelerate biological aging at the cellular level - and that clearing the accumulated damage can reverse conditions that diet changes alone cannot.

Both findings arrive at a moment when the pharmaceutical and biotech industries are rapidly converging on metabolic disease as the defining therapeutic frontier of this decade. Each one opens a pathway that existing drug development was not targeting, and each demonstrates that the body's metabolic architecture contains regulatory mechanisms that decades of research had not identified. The acceleration is not only in AI and energy infrastructure. It is in the speed at which biological systems are yielding their structural secrets to researchers equipped with tools and frameworks that did not exist five years ago.

Grid Storage Crosses a Threshold in New England

New England's largest grid battery - a 250 MW facility in Medway, Massachusetts - came fully online in late February, already surpassing the 175 MW Cross Town battery in Maine that The Century Report documented earlier this year. A 700 MW / 2.8 GWh successor at a former oil-storage site in Everett, Massachusetts is targeting completion by 2028 or 2029, which would make it among the largest stand-alone batteries in the country.

The driver behind this wave is Massachusetts' Clean Peak Standard, which requires utilities to serve growing percentages of peak-demand hours with clean electricity. The rule creates a revenue mechanism for batteries that store midday solar energy for evening dispatch - exactly when consumption spikes as people return home. After years of building small-scale batteries in the 1-5 MW range, the state's policy architecture is now producing multihundred-megawatt projects with commercial financing structures that institutional investors can underwrite.

Simultaneously, U.S. community solar crossed 10 GW of installed capacity, though the growth trajectory is complicated. Installations fell 25% in 2025 as mature markets like New York and Maine saturated. Wood Mackenzie forecasts a 12% rebound in 2026 driven by Illinois and Mid-Atlantic states, but longer-term growth depends on new state programs in Ohio, Iowa, Pennsylvania, and Michigan. Developers are sitting on more than 8 GW of projects with 29% already under construction. The sector's evolution from small, distributed installations to community-scale projects up to 20 MW reflects a broader pattern: distributed energy is scaling into grid-significant infrastructure, not through any single breakthrough, but through the steady accumulation of policy, technology, and capital flowing in the same direction. That extends the distributed-buildout arc the April 8 edition of The Century Report followed when Maine joined the balcony-solar legislative wave spreading across roughly half of U.S. states.

The Century Perspective

With a century of change unfolding in a decade, a single day looks like this: the White House prepares to place Mythos inside the departments that administer defense, finance, justice, and state power, finance ministers and central bankers treat one model as a matter for coordinated government attention, Anthropic turns Opus 4.7 into a live proving ground for safeguards before wider Mythos-class access, robots begin inferring how to act from fragments of experience they were never directly taught, metabolic science opens fresh routes to reverse obesity and fatty liver disease through mechanisms hidden in plain sight, and New England grid batteries and community solar keep expanding the hours when clean power is available. There's also friction, and it's intense - the same administration now distributing Mythos is still fighting over whether Anthropic itself constitutes a supply-chain risk, regulators and bank chiefs are openly admitting they have no settled framework for a capability already entering core institutions, community solar growth slowed as early state markets saturated, and the legal and policy architecture around frontier AI is being written in the middle of deployment rather than before it. But friction generates refraction, and refraction is how a system reveals the medium it has been passing through all along. Step back for a moment and you can see it: frontier intelligence crossing from product release into the operating layers of government and finance, biological research exposing overlooked control systems inside metabolism and aging, and energy infrastructure maturing from scattered projects into regional capacity shaped by policy design. Every transformation has a breaking point. Gravity can collapse what is unprepared to bear its pull... or gather scattered mass into worlds with enough coherence to hold life.

AI Releases & Advancements

New today

  • Anthropic: Released Claude Opus 4.7, now generally available across Claude products and the API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry. (Anthropic)
  • OpenAI: Introduced GPT-Rosalind, a frontier reasoning model for life sciences research workflows. (OpenAI)
  • OpenAI: Updated the Codex app for macOS and Windows with computer use, in-app browsing, image generation, memory, and plugins. (OpenAI)
  • Google: Released Android CLI, a terminal-first interface for Android development designed to work with AI agents. (Android Developers Blog)
  • Google: Rolled out AI Mode in Chrome updates that add side-by-side page viewing and the ability to search across recent tabs, images, and files. (Google Blog)
  • Google: Began rolling out personalized image creation in the Gemini app using Personal Intelligence, Nano Banana 2, and Google Photos context. (Google Blog)
  • Qwen: Released Qwen3.6-35B-A3B, an open-weight 35B MoE model with 3B active parameters for agentic coding. (Qwen Blog) on Hugging Face under Apache 2.0. (Hugging Face)
  • Cloudflare: Launched Artifacts in beta, a Git-compatible versioned storage system for AI agents. (Cloudflare)
  • Cloudflare: Launched Email Service in public beta for AI agents to send and receive messages. (Cloudflare)
  • Cloudflare: Released AI Platform, an inference layer for running agent workloads. (Cloudflare)
  • Canva: Launched Canva AI 2.0, an updated AI design suite with collaborative creation and external tool/workflow connections. (Product Hunt)
  • Windsurf: Released Windsurf 2.0 with an Agent Command Center and Devin integration. (Product Hunt)

Other recent releases

  • Google DeepMind: Released Gemini 3.1 Flash TTS, a text-to-speech model with audio tags, multi-speaker dialogue, and 70+ language support, now rolling out in preview via the Gemini API, Google AI Studio, Vertex AI, and Workspace. (Google DeepMind Blog)
  • OpenAI: Updated the Agents SDK with native sandbox execution and a model-native harness for building secure, long-running agents across files and tools. (OpenAI)
  • Reka: Launched Reka Edge, an edge intelligence model for physical AI applications. (Product Hunt)
  • Tencent: Released HY-World 2.0, an open-source 3D world model that generates Gaussian splats, meshes, and point clouds from text or image inputs, with Unity and Unreal export support. (Reddit)
  • Fathom: Released Fathom 3.0, adding bot-free operation and integrations with ChatGPT and Claude to its AI meeting notes platform. (Product Hunt)
  • OpenAI: Introduced GPT-5.4-Cyber through its Trusted Access for Cyber program for vetted cybersecurity defenders. (OpenAI)
  • Google Chrome: Launched Skills in Chrome, letting users save and reuse Gemini prompts as one-click workflows in the browser. (Google)
  • H Company: Released HoloTab, a Chrome extension for browser-based computer use and routine automation, available now in the Chrome Web Store. (Hugging Face Blog)
  • NVIDIA: Released NVIDIA Ising, an open family of AI models for quantum processor calibration and error-correction decoding, along with training and deployment workflows. (NVIDIA Developer Blog)
  • NVIDIA: Released the NVIDIA ALCHEMI Toolkit, a PyTorch-native set of GPU-accelerated building blocks for atomistic simulation workflows in chemistry and materials science. (NVIDIA Developer Blog)

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