Sign in Subscribe
Century Report

White House Kills AI Safety Order - TCR 05/22/26

Trump pulled a draft AI safety order hours before signing, London blocked a £50M Palantir police deal, and OpenAI disproved an 80-year-old Erdős conjecture.

Ben Linford

Ben Linford

— 18 min read
Four-panel infographic: OpenAI cracks 80-year Erdős conjecture, Spotify-Universal AI remix license, London blocks Palantir Met deal, SpaceX S-1 reveals Anthropic's $15B compute spend.

The 20-Second Scan

The 2-Minute Read

The thread across yesterday's signal is the governance layer for AI being assembled and disassembled on the industry's clock while the capability compounds on its own. A draft executive order requiring frontier model pre-deployment testing was pulled at the eleventh hour after intervention from the administration's former AI czar. A mayor blocked a procurement contract over absent due process. The institutions trying to govern frontier capability are operating with frameworks built for a slower era, while the capability itself just disproved an open conjecture that survived eight decades of mathematical attention.

The capital layer compounded in parallel. SpaceX's S-1 filing made the financial architecture of frontier AI legible for the first time at this granularity: $15 billion annually flowing from Anthropic to xAI infrastructure, $2.8 billion in gas turbines, a $530 million litigation reserve attached partly to Grok's most permissive output modes. The physical and legal substrate of the intelligence era now sits in a regulatory filing on the public record.

The consent layer extended through Spotify and Universal Music agreeing the first major-label AI remix architecture, giving top-tier catalog a working compensation pathway when AI generates derivatives from it. The friction layer arrived at graduation stages and termination notices: 2026 graduates booing CEOs who frame AI adoption as obligatory, and an Australian logistics firm sending two different termination narratives to employees in two jurisdictions in the same week.

What the cycle traces is institutional architecture forming faster than the federal-level deadlock would suggest - in courtrooms, procurement contracts, mayoral vetoes, licensing frameworks, and SEC filings - even while the top-down review framework keeps reversing itself.

The 20-Minute Deep Dive

Trump Administration Kills AI Security Executive Order Hours Before Signing

Hours before the White House was scheduled to announce signing, the administration pulled a draft executive order that would have mandated 90-day pre-deployment safety testing of frontier AI models by the NSA and Treasury Department. According to Politico's reporting, the kill shot came from David Sacks, the administration's former AI czar, after tech industry complaints that the requirement would slow US competition with China. The reversal undoes a directional shift The Century Report covered earlier this month, when the Commerce Department announced its own pre-deployment testing program — itself a departure from the hands-off posture of the original AI Action Plan.

The governance layer for frontier AI is assembling and disassembling on industry timelines rather than regulatory ones. The Commerce Department's testing announcement, the draft executive order, and the eleventh-hour reversal all happened inside roughly three weeks. The same week the order was killed, House negotiators advanced a separate bill that would preempt state AI safety laws — including New York's and California's frameworks — for two years, drawing fire from state legislators who see the federal vacuum as license for them to legislate.

The deeper read connects to the Anthropic-Pentagon arc, the Mythos coordinated-disclosure architecture, and the Five Eyes agentic AI guidance this newsletter has tracked across recent months. The institutional framework for AI capability is being built in pieces - courtroom precedents, industry-led release protocols, regulator-by-regulator preview access, cross-jurisdictional review layers - rather than through a single coherent federal architecture. The pieces are forming faster than the federal-level deadlock suggests. Every reversal of a top-down framework forces capability into the bottom-up channels that are already proving more durable: published audits, staged-release governance, judicial review of executive actions, and the verification standards being written into procurement contracts by buyers who can no longer afford to take supplier claims on faith.

The order would have routed safety testing through NSA and Treasury. Its absence routes it through what the labs publish voluntarily, and the D.C. Circuit spent last week pressing whether punishing those publications is constitutional. The 24-hour federal reversal is the visible layer. Underneath it, an audit record assembled from voluntary lab disclosures is acquiring constitutional protection on a separate docket, on cycles measured in months rather than hours.

London Blocks Its Police Force's Largest AI Procurement to Date

The Mayor of London moved this week to halt a £50 million contract between the Metropolitan Police and Palantir, citing absent public consultation and unresolved questions about how the company's analytics platform would handle data on Londoners never suspected of any crime. The decision stops what would have been one of the largest AI-related procurements in UK policing history before any data flowed through the system. As The Century Report covered on April 26, the Met had already run Palantir's analytics system across a week of internal surveillance, producing 3 arrests, 98 misconduct assessments, 500 prevention notices, and 42 senior-officer attendance-fraud reviews - operational adoption that outpaced the formal procurement oversight this decision is now seeking to restore.

The structural significance lies in the precedent. Police AI procurement in the UK has largely run through internal force decision-making and Home Office sign-off, with elected oversight applying after the fact if at all. Inserting a democratic veto upstream of deployment changes the geometry of how police-grade AI gets adopted across the country. Forces watching this decision will now have to model elected pushback as a procurement-stage variable, where it was previously treated as a post-deployment risk.

Palantir's response was muted, noting that the Met retains the option to revisit the deal under a different procurement structure. But the company's UK growth strategy has leaned on the assumption that operational urgency inside police forces would outpace civic deliberation. That assumption hit a measurable obstacle in the capital city with the country's largest police force. The further this pattern spreads, the more AI vendors will have to design for civic legibility from the outset rather than retrofitting transparency under pressure. The economics of selling opaque, all-purpose analytics into public institutions on operational urgency alone are getting more expensive on their own terms, and the cost of building for democratic legibility from day one is starting to look like the cheaper path.

OpenAI Disproves an 80-Year-Old Erdős Conjecture

In 1946 Paul Erdős posed a question about points in the plane: if you place n points so that the count of pairs at distance exactly one is maximized, what does the configuration look like? He conjectured the answer lay along a particular bound tied to triangular grid structures. Mathematicians have chipped at the problem for nearly eight decades, tightening upper and lower bounds, producing partial results, but never breaking the conjecture or proving it. A reasoning model from OpenAI just disproved it.

The construction is small enough that a mathematician can verify it by hand. The model produced an arrangement of points that exceeds the bound Erdős predicted, and the configuration is novel — it does not appear in the literature, and the expert reviewers who examined it described it as the kind of solution that requires either a leap of insight or an exhaustive search through a space far larger than unaided humans can traverse. The model appears to have done the second, which is itself a category claim worth examining. Search has always been part of mathematical discovery. The depth and breadth at which it now operates is what has changed.

This sits inside a larger pattern. The May 16 edition of The Century Report documented the predecessor: ChatGPT closing Erdős Problem 1196 on covering systems of congruences alongside Stanford mathematician Mark Sellke in roughly 80 minutes. AlphaProof solved IMO-level geometry problems. AlphaGeometry produced original proofs. AlphaTensor found faster matrix multiplication algorithms. The Erdős disproof is the first time a publicly-deployed reasoning system has overturned a long-standing open conjecture from the mainstream literature, with a construction small and clean enough that human verification was straightforward.

What this implies for the open problem book is where coverage of this result tends to stop. There are thousands of conjectures of comparable structure - small constructions that would settle them if found, search spaces too large for unaided humans, no obvious path of attack. Many are like the Erdős conjecture: posed decades ago by mathematicians who recognized the problem was hard but suspected the answer lay just outside reach. The pace at which mathematical knowledge accumulates was, until recently, set by how many of these problems a generation of researchers could keep in active working memory. That ceiling is moving. The coming years of pure mathematics will look different from the recent decades behind us, and the difference will be visible in problems closed faster than the surrounding theory can absorb the closures.

Spotify and Universal Build the First Licensed AI Remix Architecture for a Top-Tier Catalog

Spotify and Universal Music Group agreed on a licensing framework letting premium subscribers generate AI-driven covers and remixes from Universal's catalog, including songs by Taylor Swift, Ariana Grande, and Billie Eilish. The deal is expected to ship as a paid add-on through the Spotify app, with revenue flowing back to artists and songwriters. The companies framed the agreement around what they called "consent, credit, and compensation" for participating artists. Spotify shares rose 16% on the announcement, with the company guiding investors toward mid-teens revenue growth and 35-40% gross profit margins through 2030.

The structural significance: this is the first time a major streaming platform has built explicit consent and compensation infrastructure for AI-generated content using catalog from a top-tier label. The architecture extends a thread the May 13 edition of The Century Report tracked when the Human Consent Standard launched, a machine-readable licensing protocol George Clooney, Tom Hanks, Meryl Streep and a performer coalition backed for personal likeness and voice. The Academy of Motion Picture Arts and Sciences ruled on May 3 that Oscar eligibility requires performances "demonstrably performed by humans with their consent" and human-authored screenplays. Three threads of the same architecture forming across creative industries within the same six-week window.

What the Spotify-Universal deal demonstrates is that the consent layer is moving from theoretical to operational. The prior landscape was litigation without revenue: rights holders sued AI labs over training, courts issued partial rulings, and the day-to-day deployment surface for AI-generated music ran outside any compensation pathway. Listeners encountered AI impersonations of musicians on the platform faster than Spotify's detection systems could respond. Deezer recently reported 44% of new uploads were AI-generated and 97% of listeners could not distinguish them from human-made songs. The friction between AI capability and creative-industry economics was generating courtroom records but no shared infrastructure.

The deal alters that landscape by giving the catalog itself a compensation pathway when AI generates derivatives. Artists who participate receive revenue. The platform charges for the capability. The catalog earns when its constituent songs become inputs to new derivative works. The economics inverted on their own commercial terms in this corner of the industry: the path that withheld licensing and fought the technology generated lawsuits and no revenue, while the path that built a consent framework produces both revenue and a verifiable infrastructure for what AI generates from licensed inputs. The next several quarters will surface whether visual-art catalogs, audiobook narration, and journalism source material follow the same path.

Graduates Boo the Inevitability Speech as OpenAI Hires a Crisis Strategist

The wave of viral videos showing 2026 university graduates booing commencement speakers who praised AI is now a pattern rather than a single incident. Former Google CEO Eric Schmidt drew a chorus of boos at the University of Arizona after telling graduates AI is a rocket ship they should board without asking which seat. Gloria Caulfield, an executive at a property development company, received what The Verge described as an icy reception from arts and humanities graduates at the University of Central Florida. Music industry CEO Scott Borchetta gave a patronizing response to AI hecklers at Middle Tennessee State University, telling students critical of AI to "deal with it."

What the videos capture is a cohort entering the bleakest US job market in recent memory, watching corporate executives who have argued AI will reshape much of the white-collar workforce frame that reorganization as something graduates should welcome. Penny Oliver, a recent George Mason graduate, told The Verge the executives' surprise at the reaction revealed the disconnect. Austin Burkett, a recent NYU Game Center MFA graduate, noted that several of his former classmates have taken gig work training AI models that will substitute for the roles they trained for.

The pattern arrived in the same news cycle as a Wired profile of OpenAI's chief of global affairs Chris Lehane, the Clinton-era political operative who later managed Airbnb's hostile-regulator phase. The pairing reveals what is actually scarce inside the leading AI labs right now. Capital is abundant. Talent is being acquired at compensation packages that would have been unimaginable three years ago. Compute is constrained but expanding on every continent. The variable that no balance sheet expands is social license - the diffuse permission a society extends to a technology before regulation has codified it, and the absence of which makes everything from hiring to enterprise sales to policy access progressively harder.

Lehane's record at Airbnb is the closest playbook the industry has. The company faced city-by-city regulatory revolt and responded by reorganizing around community engagement, host mobilization, and proactive disclosure. The shift was that operating without explicit local consent stopped being feasible as a default; the company had to earn renewal in each market continuously. AI labs are converging toward a similar position. The bargaining infrastructure is forming. The labs that recognize this early and build for legitimacy as a continuous earned resource will operate on different ground than the ones still pricing social license at zero.

WiseTech Cuts 2,000 Jobs for AI Reasons, Omits AI from Chinese Employees' Notices

WiseTech Global, the Australian logistics software firm whose CargoWise platform handles a large share of global freight forwarding, began cutting roughly 2,000 positions this week. The Century Report covered on May 7 what preceded these cuts: WiseTech staff left in limbo after management touted AI as "better than humans," with workers describing their craft as structurally obsolete. The English-language internal communications and the company's investor messaging name the cause directly: AI-driven productivity gains have made entire job categories redundant. The termination notices sent to the company's Chinese employees, reviewed by the Guardian, contain no mention of AI. They cite "business restructuring" and "operational realignment."

The omission is unlikely to be accidental. China's labor law framework requires companies to negotiate restructuring with employee representative bodies and provides for compensation calculations that depend on the stated cause of termination. Citing AI as the reason - particularly in a jurisdiction where AI-driven workforce reduction has not been tested in court - exposes the company to legal contests that simply listing "restructuring" sidesteps. Two parallel termination narratives are now circulating inside the same company, on the same week, for the same underlying reason, and the version each employee receives depends on which legal system governs their employment contract.

This is the early shape of what AI-driven workforce reduction looks like when it meets jurisdictionally fragmented labor regimes. The companies running the reductions need legal regimes to treat the cause as either disclosable or undisclosable, and right now the regimes do not agree. In Australia, where union pressure and parliamentary committees have spent recent months asking employers to be explicit about AI-driven cuts, naming the cause has become standard. In China, where the framework around AI-and-labor has not been written, omitting the cause is the path of least resistance.

What the WiseTech case shows is that the labor-disclosure regime for AI-driven reductions is being built, week by week, by the companies doing the reducing and by the regulators reacting to what they see. A company that names AI in one termination notice and omits it in another is producing the evidence that future labor frameworks will be written against. Worker advocacy groups in both jurisdictions now have the same internal document set; the comparison is already being made publicly. The disclosure asymmetry will close - either through legislation that requires uniform disclosure across jurisdictions, or through worker action that makes the omission as costly as the disclosure. The pattern visible in this week's notices is the friction of that asymmetry resolving, and the resolution is moving toward more disclosure rather than less.

The same internal document set is now in the hands of worker advocacy groups in both jurisdictions, sourced and translated and published. A company optimizing for the cheapest per-jurisdiction disclosure is the mechanism by which the cheapest disclosure stops being available. The Guardian's publication of both versions is the moment the arbitrage stops paying on its own commercial terms, and the labor frameworks that catch up next will be written against the comparison this week put on the record.

The Other Side

Until this spring the vocabulary for refusing the "AI is inevitable, board the rocket" framing did not exist in commercial speech. Graduates were told the technology was the weather. The lines worked at previous commencements because the audience had no compact phrase for what was missing from them. The boos now arriving at the University of Arizona, the University of Central Florida, and Middle Tennessee State are the audience supplying the phrase.

Austin Burkett's note that his former NYU Game Center classmates are training the AI models that will substitute for the design work they spent two years learning is a sentence that did not have a home in commercial vocabulary six months ago. It does now. The cohort booing on stage is the same cohort writing the words the next CEO speech will have to answer in.

OpenAI's hire of Chris Lehane is the inside of the same trade. Capital is abundant. Talent is bid up. Compute is constrained but expandable. The variable no balance sheet expands is the permission a society extends to a technology before regulation requires it, and that permission just stopped extending itself in three commencement halls in the same news cycle. Lehane's Airbnb resume is what labs reach for when operating without continuous local consent stops being free. What they are acquiring is the architecture for negotiating with publics whose vocabulary has caught up.

At this moment Burkett's former classmates are at screens labeling the data that will train the models for the design jobs they no longer have. The next time Lehane drafts a commencement remarks package for a frontier-lab CEO, he will have to write around the room those classmates are in.

The Century Perspective

With a century of change unfolding in a decade, a single day looks like this: a reasoning system overturning an open conjecture that survived eighty years of unaided human attention, a top-tier music catalog acquiring the first working consent architecture for AI-generated derivatives, the financial substrate of frontier compute becoming legible in a single regulatory filing at unprecedented granularity, a coalition of Western governors locking in geothermal procurement at gigawatt scale, the verification layer for AI capability assembling in courtroom precedents and procurement vetoes the federal deadlock did not predict. There's also friction, and it's intense - a pre-deployment safety order killed hours before signing after industry pressure on the administration's former AI czar, a £50 million police-Palantir contract blocked by a mayor over absent public consultation, an employer producing two different termination narratives across two jurisdictions in the same week, graduating cohorts loudly booing the CEOs who frame AI adoption as obligation, researchers who study online disinformation suing a State Department whose visa restrictions now reach the people studying the hate speech. But friction generates traction, and traction is what lets a force finally grip the surface it has been spinning against. Step back for a moment and you can see it: the consent infrastructure for creative work taking shape while the underlying litigation was still being argued, the substrate of intelligence economics becoming readable to anyone with an SEC subscription, the pace at which open mathematical conjectures fall compressing into spans the prior generation had measured in decades, the bargaining infrastructure for what AI deployment costs assembling at graduation stages and mayoral offices and labor courts simultaneously. Every transformation has a breaking point. A wedge can split the structure it enters... or hold open the seam that everything afterward depends on passing through.

AI Releases & Advancements

New today

  • Tencent: Open-sourced Hy-MT2, a family of "fast-thinking" multilingual translation models in three sizes (1.8B, 7B, and 30B-A3B MoE) supporting translation across 33 languages; the 7B and 30B-A3B models outperform DeepSeek-V4-Pro on translation benchmarks in fast-thinking mode, and the 1.8B can be quantized to 440MB via AngelSlim 1.25-bit; also releases IFMTBench, a new translation instruction-following benchmark. (GitHub)
  • NuMind: Released NuExtract3, an open-weight 4B vision-language model for structured information extraction and OCR under Apache 2.0; accepts text, images, or mixed inputs and outputs structured JSON or Markdown, supporting invoices, receipts, forms, contracts, and scanned documents with both reasoning and non-reasoning inference modes. (Hugging Face)
  • Runtime (YC P26): Launched a sandboxed coding agent infrastructure platform enabling engineering teams to run Claude Code, Codex, and other AI coding agents in isolated environments without custom DevOps setup. (Runtime)

Other recent releases

  • Microsoft: Open-sourced RAMPART and Clarity, two AI agent safety tools for developers; RAMPART embeds adversarial and benign safety tests into CI pipelines for agentic AI systems, while Clarity is a structured pre-build review tool to surface misaligned assumptions before coding begins. (Microsoft Security Blog)
  • NVIDIA: Released NVIDIA Verified Agent Skills, a framework for cataloging, scanning, signing, and documenting portable AI agent skill packages; includes SkillSpector, an open-source scanning tool that checks agent skills for prompt injection, tool poisoning, trigger abuse, and supply-chain risks before publication to the NVIDIA/skills GitHub catalog. (NVIDIA Developer Blog)
  • Google DeepMind: Launched Gemini for Science in early access via Google Labs, a suite of AI research tools including Hypothesis Generation (analyzing millions of papers for hypothesis support), Computational Discovery (an agentic search engine for running experiments), Literature Insights (chat-based literature review), and Science Skills (integrating 30+ life science databases including AlphaFold and UniProt). (Google Blog)
  • 1Password: Launched a Trusted Access Layer integration for OpenAI Codex that gives AI coding agents access to enterprise credentials during development workflows without exposing secrets in prompts, source code, repositories, or terminal output. (1Password Blog)
  • Google DeepMind: Released Gemini 3.5 Flash, the first model in the new 3.5 series, delivering frontier-level coding and agentic performance at 4× the speed of comparable models; outperforms Gemini 3.1 Pro on Terminal-Bench 2.1 (76.2%), MCP Atlas (83.6%), and CharXiv Reasoning (84.2%); available now globally in the Gemini app, AI Mode in Search, Google Antigravity, and the Gemini API. (Google DeepMind)
  • Google DeepMind: Released Gemini Omni, a new natively multimodal model that generates any output from any input starting with video; combines Gemini reasoning with Veo, Nano Banana, and Genie for video understanding, editing, and generation; rolling out now to Google AI Plus, Pro, and Ultra subscribers via the Gemini app and Google Flow. (Google DeepMind)
  • Google: Launched Antigravity 2.0 at Google I/O, repositioning the agent-first development platform with parallel multi-agent orchestration, new CLI tools, SDK, voice support, and integrations with Firebase, Android Studio, and AI Studio; available now for developers. (Google Developer Blog)
  • OpenAI: Launched support for Google's SynthID watermarking in GPT image outputs and released a new AI content provenance verification tool, enabling users to check whether images were generated by AI; both available now. (OpenAI)
  • Allen Institute for AI (Ai2): Released OlmoEarth v1.1, a new family of remote sensing foundation models that cut compute costs by up to 3× versus OlmoEarth v1 while maintaining comparable performance on satellite imagery tasks including crop-type mapping and forest-loss classification; available on Hugging Face. (Hugging Face Blog)
  • xAI: Enabled Grok for use inside OpenClaw, an open-source local-first AI agent, allowing SuperGrok and X Premium subscribers to run Grok within the OpenClaw desktop agent. (xAI)
  • JHU CLSP / Sentence Transformers: Released the Ettin Reranker Family, six open CrossEncoder rerankers (17M–1B parameters) built on ModernBERT encoders and trained via distillation, setting state-of-the-art performance at each respective size on MTEB Retrieval; all support 8K-token context and are available on Hugging Face. (Hugging Face 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