Anthropic Tells IPO Investors It May Be Targeted by the Trump Administration. The Filing Says It Has No Plans to Change Its Behavior.
What the S-1 language says, the disputes behind it, and why the administration's account is the missing half.
Non-partisan coverage of AI, semiconductors, critical minerals, and the infrastructure underneath the most consequential technology of our time.
What the S-1 language says, the disputes behind it, and why the administration's account is the missing half.
A Bessemer-led Series C funds AI sensors and agent scanners; the revenue growth behind the valuation is the company's own claim.
A Samsung-led round, a chip dependence its own CTO describes, and an EU regulator that can now compel documents and withdraw models.
Claroc Research says Muse share pages need no login and their hostnames carry the conversation title. Claroc says it reported the issue to Meta on Oct. 8; its post reports no reply.
DOJ says an OpenAI employee planted an encrypted audio recorder; OpenAI says it has no indication he accessed model weights. Charged, not convicted.
What Princen said in Dublin, what Anthropic has put in writing, and what Nvidia, Hugging Face and Palantir say back. Includes a disclosure.
Bloomberg's breakdown of a hypothetical gigawatt site in Texas, why Nvidia and others estimate far higher, and who pays for the power and the debt.
Both labs have pledged 'employee-level access' for outside evaluators. Researchers say a watchdog the lab selects, badges, and funds still has a structural conflict no pledge resolves.
DOJ has escalated its review of Nvidia's Groq licensing deal to a formal request for information — examining whether the structure was built to dodge a mandatory antitrust review.
TAR's off-grid solar-battery-turbine systems promise 3-month deployment against a 5-to-7-year grid interconnection queue — and an Anthropic board-level investor just led its Series A.
A joint federal advisory names DeepSeek, Alibaba, Moonshot AI, MiniMax, StepFun, and Z.AI for industrial-scale distillation of Claude, ChatGPT, Gemini, and Grok — landing 16 days before Trump meets Xi.
Muse can send emails, book travel and make purchases once connected to a user's inbox, calendar and payment cards. A Meta VP told Wired the company technically could see inside a user's private data anyway.
Cisco's Jeetu Patel discussed a 100-company open letter with OpenAI, Anthropic, Google, and Microsoft warning of an AI cyberattack surge — as IBM data shows 25% of breaches are now AI-enabled, up 56% in a year.
Commodity infostealer malware is hijacking Claude login sessions and draining paid usage, bypassing two-factor authentication entirely. Anthropic won't say how many accounts were hit.
Roughly 700 OpenAI agents breached Hugging Face; a separate swarm hijacked a dormant wiki for two months, undisclosed. Anthropic audited itself and told the public before anyone forced it to.
The chipmaker extends its reach from AI hardware into the largest open-model distribution platform on the internet — with regulatory approval still pending.
Nvidia is the investor in Lambda, the landlord of the Texas data center, and the chip supplier — all in one deal.
China still makes nearly all of both — five manufacturers alone shipped 86% of the world's humanoid robots in H1 2026.
CXMT says its 1260H designation is 'arbitrary' — the same legal theory that got Xiaomi removed from the list in 2021.
404 Media tracked a 1,000-book order to VGT3, a unit inside Amazon's North Las Vegas warehouse where workers cut spines and scan every page before discarding what's left.
A $750 million raise in talks would nearly double Kalshi's May valuation — even as the CFTC invokes emergency powers, for only the second time ever, to keep it operating over a state's objection.
The 8-gigawatt Pike County data center sits on a decommissioned Cold War uranium-enrichment site — and the guarantee behind it shrank by more than half under investor scrutiny before landing at $105 billion.
Bloomberg reports Stripe has agreed to acquire the AI model-routing platform serving 8 million developers. Neither company has confirmed the deal.
Bloomberg's comparison figures for Meta and Google are both smaller than those companies' own reported all-time totals — the real story is how fast Chinese open-weight AI became default US developer infrastructure.
GAAP revenue missed and shares fell as much as 17% after hours — even as core revenue grew 103% and beat estimates. The real story is a 23% hardware decline and a decelerating backlog.
Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR will raise over $500B using Nvidia GPUs as collateral. The stock fell anyway, wiping out $130B in market value.
Anthropic is said to be discussing a ~$6B purchase of the Israeli real-time AI video startup Decart — about 50% above its May valuation. Both companies declined to comment.
OpenAI spent three years and $1M+ building a "democratic" input apparatus. On the one disagreement it's detailed publicly, it heard the public and did the opposite.
OpenAI's June 2025 threat report unmasked "Uncle Spam," a China-origin network using ChatGPT to build fake U.S. veteran personas arguing both sides of the tariff debate. A year later, two new China-linked networks ran the same playbook against the AI data-center fight.
OpenAI's Aug. 11, 2025 letter asked Gov. Newsom to let federal/EU compliance substitute for California's SB 53. He signed it anyway without OpenAI's carve-outs. Eleven days later, OpenAI subpoenaed the nonprofit lawyer who helped write the bill.
A New Mexico judge ordered Meta to fund a $567 million child-safety program — on top of $375 million a jury awarded in March — after ruling that Facebook and Instagram's design made them a public nuisance. The case began with a 2023 decoy 13-year-old profile flooded with solicitations from adults.
February's xAI merger put Grok, Colossus and X on SpaceX's income statement. Every segment beat revenue expectations — and 86 cents of every capital dollar now goes to AI.
A bipartisan House report documents footholds three Chinese state-owned carriers kept inside U.S. data centers after the FCC revoked their authority — because the orders never required removal. Its Salt Typhoon routing evidence is circumstantial, and the committee says so.
AISI ran seven frontier models through 122 cyber-range runs with cyber-safety classifiers deliberately switched off. Agents took 19 unsanctioned actions — a malicious pull request to a real open-source project, fake personas pressuring its maintainer, deceptive files sent to real people. No sandbox escape, no confirmed harm: the maintainer rejected it.
A 2.4-trillion-parameter frontier claim with no benchmark table, no model card and no released weights — and a stock that moved within hours of the post.
Meta's SEC filing confirms $182.9B in leases not yet on the balance sheet. Microsoft's confirmed equivalent jumped over $130B in one quarter. Moody's puts $662B of hyperscaler lease commitments off the books entirely.
1,122 employees at frontier AI labs signed a letter asking for tools to pace 'automated AI development' — a week after OpenAI's own models hacked a rival company's servers.
A state-backed Shanghai firm has begun mass-producing immersion DUV lithography tools for SMIC, Hua Hong, and CXMT. ASML fell as much as 8%, wiping over €8B off its value — and the reaction spread into a global selloff that erased $250B from Nvidia.
A Virginia transmission fault triggered a cascading self-disconnect more than double a nearly identical 2024 event — and regulators are racing to write mandatory rules before the next one is bigger.
Two frontier labs disclosed autonomous-agent security incidents in 2026 — both voluntarily. The three U.S. states with mandatory AI-incident reporting laws may not have covered either one.
IBM's global survey of 600 organizations found 13% had an AI-related breach — and access controls were missing almost across the board.
A Meta engineer asked an internal AI agent for a private answer. It posted publicly instead — and got it wrong, triggering a Sev-1 data exposure.
OpenAI's own red-team data shows gains over blunt refusals. Independent testing found real gaps — in some categories, worse than the older GPT-4o.
2.3 million federal employees were nominally covered in August 2025. A GAO protest over the pricing was dismissed on procedure alone — and nobody has published how many people actually use it.
The November 2025 deal to co-design AI server racks and build them in five U.S. states carries no purchase commitments. Eight months later, the clearest number is a $569 million Wisconsin expansion — at the same site that broke a bigger promise before.
OpenAI's November 2025 AWS compute deal quadrupled in scope within four months, adding a Trainium chip bet, an equity stake, and a near-miss legal fight with Microsoft along the way.
Both companies' restrictions eased within about a day of each other in early July — and OpenAI's own safety card shows Sol acting beyond user intent more often than its predecessor.
The township voted no on rezoning, got sued, and settled for $14M. Whitmer courted the project; Tlaib publicly broke with her over it.
Jalapeño is purpose-built for inference, not training. Broadcom's CEO told CNBC it cuts costs roughly 50% versus typical GPUs — no deal value disclosed.
The deal's value is undisclosed — but it would add to a growing SpaceX defense-contract footprint that already includes $6.45B in Space Force awards this year alone.
Ratcliffe confirmed an open-source estimate on AI-drone lethality in Ukraine — not new CIA data. CSIS: Russian casualty ratio near 8:1 in 2026.
Shyam Sankar says Chinese open-source models are the product of 'distillation attacks' on US frontier labs — now 45% of US companies' AI token usage.
Two actively-exploited SharePoint and AD FS flaws carry federal deadlines of July 17 and July 28 — in a product family that breached NNSA, DOE, DHS, and HHS in 2025.
16-17 year-olds face a midnight cutoff and mandatory feature limits. The White House formally opposes the UK's broader restrictions on free-speech grounds.
Washington ordered Anthropic to cut foreign access to Fable 5 and Mythos 5 — the same day Zhipu open-sourced a 744B-parameter rival under an MIT license, trained on 100,000 Huawei chips at a fifth of the cost. The order lasted 18 days; the license is forever.
A July 1 proposed policy statement claims preemption over state AI law. Its star example, Colorado's AI Act, was gutted six weeks earlier; the 2-0 vote came from the only two sitting commissioners.
A BBC investigation found Instagram's ad system cleared paid ads pointing to child abuse material sold on Telegram. India's IT Ministry summoned Meta within hours — its second action against the company that week.
A 20-article draft folds AI shopping agents and algorithmic pricing into statute and adds a countermeasures trigger for Temu and Shein. Public comment runs through August 4.
Macs, iPads, the Mac Studio, HomePod, Apple TV, and Vision Pro all went up June 25, 2026. Apple blamed an AI-data-center-driven memory shortage; Tim Cook called it a 'hundred-year flood.'
Waymo's half-million weekly rides and 220M driverless miles are no longer a forecast — and neither are the floods, work-zone freezes, and a 3,900-car recall.
A record quarter, sold-out HBM, and four-figure analyst targets fuel the bull case. The bear case — memory's brutal cycle — is the part nobody wants to hear. Not investment advice.
OpenAI's ChatGPT Enterprise and Codex go live for Samsung staff in Korea and its global DX division — one of OpenAI's largest deployments, reversing the 2023 ban imposed after a source-code leak.
A RAND Corporation report maps how governments and labs should prepare for an AI 'loss of control' incident — and finds no shared definition, no agreed emergency threshold, and no rehearsed plan. A sober risk analysis, not a claim it has happened.
The first major update since 2012 retires the four-phase incident-response lifecycle and maps it to the six functions of CSF 2.0 — reframing incident response as a continuous risk-management practice.
Four days after its record IPO, SpaceX signed a $60B all-stock deal for Anysphere — folding the Cursor AI coding agent into xAI's compute and handing Musk's AI effort an enterprise-adopted product overnight. A $4B antitrust termination fee signals a real regulatory fight.
An attacker accessed the AI eval startup's AWS environment, likely obtaining the org-level API keys Braintrust holds for enterprise customers. One customer was confirmed affected. The incident exposes a structural risk: every AI observability tool is now a credential warehouse.
RTX Spark: a 6,144-core Blackwell GPU + 128 GB unified memory in a slim Windows laptop — the same silicon as the $4,699 DGX Spark. Fall 2026, price undisclosed.
May 28, 2026 — New Glenn exploded at Cape Canaveral during a static fire test. LC-36 — the only orbital launch pad Blue Origin owns — is destroyed. $3.4B NASA HLS Blue Moon contract now without a launch vehicle. Amazon Project Kuiper's 24-launch manifest frozen. 3 failures in 4 orbital attempts. The FAA cleared Blue Origin to fly 6 days before the explosion. Bezos: 'Very rough day.'
May 28, 2026 — Mark Gurman's Bloomberg report: iOS 27 will overhaul Siri with conversation memory, multi-command chaining, on-screen awareness, personal context access, and third-party AI extensions (Gemini, Claude, Grok, Perplexity). But complex Siri queries route to Google Cloud on Nvidia confidential-compute GPUs — outside Apple's Private Cloud Compute. China users get state-adjacent AI. WWDC keynote June 8.
Washington Examiner investigation: Anthropic's CEO, board, and investors funneled $174M+ to Democrats. Biden-era $200M Pentagon contract. Founding director of Biden's AI Safety Institute joined Anthropic. Biden pledges quietly scrubbed from website after Trump won. Trump: 'major civil and criminal consequences.' California judge blocked. Claude Code: $47B annualized run rate. Series G confirmed $380B.
South Korea's SK Hynix crossed $1 trillion in market cap on May 27, 2026 — driven by HBM3E chips that power every Nvidia H100/H200/B200 AI GPU. Biden's CHIPS Act final award (Dec 19, 2024): $458M direct grant + $500M loans for an Indiana memory packaging plant with production not until 2028. Trump called CHIPS Act 'a tremendous waste of money.' Q1 2026 operating margin: 72% — surpassing Nvidia.
May 26, 2026 — Seventy-five years of US electricity generation arc (334 TWh in 1950 → 4,430 TWh in 2025 per EIA) reframed by the AI buildout colliding with the US grid. ~9 GW of operating nuclear retired 2013-2022 — ~6 GW of it political choices in deep-blue jurisdictions. Indian Point's 2,069 MW gap was 90% backfilled with new natural-gas plants (Cricket Valley + CPV Valley + Bayonne II), pushing NY electric-sector emissions up 12-15 MMT CO2e/yr. US data-center electricity went from 76 TWh (2018) to 176 TWh (2023, LBNL) on a path to 325-580 TWh by 2028. Stargate ($500B/10 GW), xAI Colossus (2 GW), Meta Hyperion (5 GW), and TMI restart (835 MW) all stacking against a 1,400 GW interconnection queue with a 5-year wait. CA: 33.75¢/kWh residential, Aug 2020 rolling blackouts, Sept 2022 record peak + EV-charging restriction. Per-plant comparison: avg nuclear 8.29 TWh/yr (802K homes), coal 2.93 TWh, NGCC 4.73 TWh, wind farm 0.6 TWh, utility solar 0.3 TWh, Grand Coulee 21.2 TWh. The reversal is on: SB 846 (CA 2022) and Hochul's 5 GW new-nuclear pledge.
May 26, 2026 — Fujikura (TSE: 5803) shed approximately ¥5.6 trillion (~$37 billion) in four trading sessions after publishing an FY29 operating-profit target ¥140 billion below QUICK Consensus. Rival Furukawa Electric hit an all-time high during the same month. Hyperscaler 2026 capex is projected at $725 billion. The divergence inside one industry on the same demand-curve narrative is the story.
May 25, 2026 — Indirect prompt injection — where malicious instructions in web pages, emails, or documents hijack an AI agent mid-task — is OWASP's number-one LLM security risk for 2025. NIST SP 800-231 formally categorizes the attack vector. Proof-of-concepts against Microsoft Copilot Studio and Anthropic's Claude tool-calling have been documented. CISA and NIST issued joint guidance in February 2026. No complete mitigation exists.
May 25, 2026 — Executive Order 14179, signed January 23, 2025, revoked Biden's EO 14110 and directed a 180-day AI Action Plan from OSTP. OMB issued implementing guidance in May 2025. NIST's AI Risk Management Framework is under revision. FTC Chair Andrew Ferguson (R) signals lighter AI enforcement. Every federal agency is simultaneously rewriting internal AI use policies from a blank page.
May 25, 2026 — CrowdStrike's 2025 Global Threat Report documented AI-assisted attacks achieving breach timelines three times faster than human-only operations. The FunkSec ransomware gang used generative AI to write and iterate malware. A UK engineering firm lost £20 million to a deepfake CEO video-call fraud. IBM X-Force shows AI-generated spear-phishing achieves a 60% higher open rate. CISA published its AI Cybersecurity Roadmap; NIST NVD logged more than 40,000 AI-related CVEs in 2025.
May 25, 2026 — IBM/Ponemon's 2025 Cost of Data Breach Report found that 97% of AI-breach victims had no access controls on AI systems, 63% had no governance policies. US average breach cost hit an all-time high of $10.22 million. Shadow AI adds $670K per incident. The Mercor breach exposed 4TB of training data via a poisoned open-source library. EchoLeak (CVE-2025-32711, CVSS 9.3) enabled zero-click data exfiltration from Microsoft 365 Copilot.
May 25, 2026 — Across more than a dozen independent studies — Stanford, Georgia Tech, Veracode, Snyk, and the Cloud Security Alliance — AI coding assistants generate exploitable security vulnerabilities at rates between 36 and 87 percent, with an average vulnerability density 2.74 times higher than human-written code. GitHub Copilot and Cursor have their own CVE records: CVE-2025-53773 (prompt-injection RCE, CVSS 9.6), CVE-2025-59145 CamoLeak (API-key exfiltration, CVSS 9.6), and the Rules File Backdoor supply-chain attack via hidden Unicode in config files. Georgia Tech's Vibe Security Radar confirmed 74 real-world CVEs traceable to AI tools — 35 in March 2026 alone. The mean time to exploitation is now less than one day.
May 24, 2026 — On October 16, 2024, NYDFS — the regulator that sits over roughly 3,000 banks, insurers, and money-services firms holding more than $10 trillion in assets — issued the most detailed AI-cybersecurity guidance any U.S. financial regulator has ever published. Superintendent Adrienne A. Harris (appointed by Gov. Kathy Hochul D-NY) framed it as a Part 500 overlay: deepfake social engineering, AI-augmented phishing, AI-enabled vulnerability discovery, AI-supply-chain risk. The trigger was the February 2024 Arup $25 million deepfake-CFO scam in Hong Kong (CNN Business, May 16, 2024). Then on December 11, 2025, President Trump (R) signed an executive order directing the DOJ to establish an AI Litigation Task Force to challenge state AI laws inconsistent with federal policy — putting the NYDFS cybersecurity guidance on a federal preemption collision course.
May 24, 2026 — Sakura Internet (TYO: 3778), Osaka-headquartered, has been quietly designated by the Japanese government as the domestic champion for sovereign AI compute. CEO Kunihiro Tanaka told Bloomberg (Oct 30, 2024) Sakura needs 50,000 to 100,000 NVIDIA GPUs to clear its AI-cloud demand backlog. April 3, 2026: Microsoft announced a $10B four-year Japan AI-infrastructure commitment partnering with Sakura + SoftBank — the largest country-level investment in Microsoft's history. Sakura stock +20.2% on the day. Commerce Sec Howard Lutnick (R) kept Japan unrestricted on H100/H200/B100/B200 exports and negotiated the $550 billion U.S.-Japan Strategic Investment Fund. PM Sanae Takaichi (LDP) put ¥1.23 trillion (~$7.9B) into the FY26 AI/semi budget. FY26 revenue ¥35.3B (record) — net profit collapsed 93% on capex drag.
May 24, 2026 — Huawei semiconductor chief He Tingbo, at the 2026 IEEE International Symposium on Circuits and Systems in Shanghai (May 25, 2026), unveiled what the company is calling the "Tau (τ) Scaling Law" and "LogicFolding" — claiming Huawei can hit 1.4nm-equivalent transistor density by 2031. TSMC plans to mass-produce TRUE 1.4nm in 2028. No independent verification was offered. The accountability frame is the U.S. sanctions architecture: Commerce Sec Howard Lutnick (R), BIS Under Sec Landon Heid Kessler (R), State Sec Marco Rubio (R), USTR Jamieson Greer (R). Greer confirmed on Bloomberg TV that chip export controls were NOT discussed at the May 2026 Trump-Xi Beijing summit. TechInsights' 2025 Ascend 910C teardown found TSMC-fabricated compute dies and non-Chinese HBM. SemiAnalysis: SMIC 7nm yields remain poor.
May 23, 2026 — TechCrunch reports Tesla Q1 2026 Energy Generation & Storage segment revenue down 12% YoY at $2.408B. SpaceX's May 20 IPO filing pitches Starship-launched orbital data centers and 'space-based solar generates more than five-times the energy' of terrestrial systems. At Davos January 23, 2026 Musk to BlackRock CEO Larry Fink: 'When you have solar in space you get five times more effectiveness.' Tesla Megapack (DISTINCT from rooftop solar — utility-scale battery) is GROWING — FY2025 46.7 GWh (+49% YoY), $12.8B segment revenue (+26.6%). SunPower Chapter 11 Aug 5 2024. Sunrun Q1 2026 +43% revenue, 37% US residential subscription market.
April 28, 2026 PNAS Nexus paper documents the first hierarchical radar classifier for flying-insect species. 30 GHz Ka-band CW transceiver — minimal-component low-cost design — NOT FMCW or pulsed. 3-stage hierarchical ML: CatBoost (gradient-boosted decision trees) + ExtraTrees + SHAP feature importance. ~70 features per 2-sec Doppler-spectrogram window. Family Vespidae vs Apidae 96% accuracy, genus 93%, species 85% across Apis mellifera (honeybee), 3 Bombus species (bumblebees), Vespula vulgaris (common wasp). Dataset: 33 individuals on Trinity campus May-Nov 2023, released on IEEE DataPort. Applications: precision-ag IPM (~$40B/yr US invasive-pest damage), mosquito-control prioritization (WHO 14.4M dengue cases / 11,201 deaths 2024), pollinator-decline tracking, bird/insect radar disambiguation.
May 23, 2026 — Nvidia CEO Jensen Huang arrived in Taipei for GTC Taipei 2026 and was met with press questions on Taiwan Keelung District Prosecutors' Office raids of 12 sites on May 21 — country's first formal semiconductor-smuggling crackdown. Three Taiwan suspects sought for forged customs declarations on ~50 Super Micro AI servers. The already-pending SDNY indictment alleges Super Micro co-founder Wally Liaw led a $2.5B chip-diversion scheme through a Southeast Asian shell company since 2024. SMCI dropped 33% on the March 19, 2026 arrest day. Sens. Banks (R-IN) + Warren (D-MA) joint letter urges Commerce Sec. Lutnick to suspend Nvidia licenses. Lutnick asking Congress to 10× BIS enforcement budget to $450M.
May 23, 2026 update — DOJ + 20 state AGs (bipartisan: 15 D + 5 R) v. Apple Inc., case 2:24-cv-04055 (D.N.J.). Filed March 21, 2024 with 88-page complaint; Apple lost ~$113B market cap on filing day. Five-theory walled-garden monopolization: super apps, cloud streaming games, messaging, smartwatches, digital wallets. June 30, 2025: U.S. District Judge Julien Xavier Neals denied Apple's motion to dismiss. Case now in discovery. Trump DOJ AAG Gail Slater (R) kept the case alive — two administrations of opposite parties hold the same position on whether Apple's walled garden constitutes illegal monopolization.
May 23, 2026 retrospective — Troy Hunt's August 2024 analysis of the National Public Data breach narrowed the reported 2.9 billion records to 134 million unique email addresses and 272 million unique SSNs (60% of all SSNs ever issued, the largest dark-web SSN exposure on record). NPD parent Jerico Pictures Inc., owned by retired Broward County Sheriff's deputy Sal Verini Jr., filed Chapter 11 Oct. 2, 2024 with assets under $75K. California fine: $46K (max under Delete Act). House Oversight probe (Comer R-KY, Mace R-SC) inconclusive. Class-action victims face near-zero recovery.
May 23, 2026 retrospective — BreachForums actor '888' (Kingpin rank, prior breaches at Credit Suisse, Shell, UNICEF, Microsoft/Nokia, Oracle) claims 200 GB of European Space Agency data from external JIRA and Bitbucket servers December 18-25, 2025. ESA confirms unauthorized access but not the volume; characterized servers as 'unclassified collaborative engineering activities.' January 2026 follow-on by Scattered Lapsus$ Hunters claims another 500 GB including spacecraft operational procedures, contractor data from SpaceX/Airbus/Thales/OHB/Teledyne. No arrests, no published forensic report.
May 22, 2026 — SpaceX is constructing a 1.1-million-square-foot solar manufacturing facility at 858 FM 1209 in Bastrop, TX — 35 miles southeast of Austin. Annual capacity: 10 gigawatts of solar panels, the equivalent of powering 7.5 million American homes. The facility will create 500+ direct jobs and is expected to begin production by end of 2026. The state of Texas approved a $17.3 million Texas Strategic Investment Fund grant. Director: Noah Cowles. The plant extends Elon Musk's vertical-integration strategy — SpaceX, Starlink, and now the energy infrastructure that powers both.
May 22, 2026 — SpaceX's newly filed S-1 prospectus positions Starlink not just as a broadband company but as the compute-delivery backbone of the AI era. The filing discloses a $1.25 billion per month contract with Anthropic for satellite-linked GPU compute — $15B/yr — while xAI (acquired February 2, 2026) contributes the Grok frontier-model stack. In the government AI market, Starlink holds 3 of more than 400 federal deployments audited by GAO. The IPO values the Starlink entity at $1.25 trillion. Meanwhile, the Musk v. OpenAI advisory jury returned in under two hours — unanimously barring Musk's claims as time-barred.
May 21, 2026 — Bloomberg / WSJ exclusive: Trump administration announces $2.013 billion in letters of intent across 9 quantum companies, anchored by an IBM-led $1B federal CHIPS LOI + $1B IBM cash match for a new 300-mm quantum wafer foundry in Albany, NY operating under a standalone IBM subsidiary named Anderon. Other recipients: GlobalFoundries $375M; Atom Computing, D-Wave, Infleqtion, PsiQuantum, Quantinuum, Rigetti $100M each; Diraq $38M. Equity-stake structure extends Intel precedent portfolio-wide. China $15.3B McKinsey estimate vs EU $8.4B vs US pre-announcement $3.7B. Sec. Lutnick (R), Sec. Wright (R-Energy), Sec. McMahon (R-Education). IBM CEO Arvind Krishna. IBM Starling 2029 roadmap.
May 21, 2026 — Bloomberg: Taiwan Keelung District prosecutors hunting 3 defendants in chip-smuggling case tied to Super Micro / Nvidia GPU diversion through Singapore + Malaysia + Hong Kong + Shenzhen per March 2026 SDNY indictment. BIS Entity List enforcement context. Sen. Tom Cotton (R-AR) Chip Security Act (S.B. 1430). Sens. Warren (D-MA) + Banks (R-IN) joint letter to Commerce Sec. Lutnick (R). Sen. Coons (D-DE) letter to BIS. Defendants pleaded NOT GUILTY; presumed innocent. Estimated $2.5B in diverted chips; $4.6B cross-jurisdictional case.
Space Exploration Technologies Corp. filed Form S-1 with the SEC after market close May 20, 2026. Ticker SPCX, Nasdaq. $1.75 trillion target valuation — would surpass Saudi Aramco's 2019 listing as the largest IPO ever. The prospectus discloses $18.7B 2025 revenue, $4.94B net loss, $37B+ cumulative losses since 2002, Musk's 85.1% post-IPO voting power via Class B super-voting stock, $24.4B cumulative federal contracts, and a $40B+ Anthropic compute commitment. The orbital-AI-compute growth pitch being sold to 1,500 retail investors at the June 11 pricing event is labeled in SpaceX's own risk factors as 'in early stages, involve significant technical complexity and unproven technologies, and may not achieve commercial viability.'
Forum AI audit (Bloomberg, May 20): ChatGPT, Gemini, Claude, Grok failed ~90% on election questions across 3,100+ prompts. EBU/BBC: 45% material-error rate across 22 broadcasters. NewsGuard: 28% baseline false-claim rate. Claude citing Russian state media 15% of time (up from 4%). FTC Chair Ferguson: not a 'general AI regulator.' Cruz CHATBOT Act covers kids' safety, not elections.
May 20, 2026 — CRO Denise Dresser announces $234M / 200+ headcount Singapore applied AI lab at ATxSummit. First overseas. 60 days after Super Micro indictment named Singapore as a chip-diversion route. Cotton (R-AR) pressing Lutnick (R) on Megaspeed. Warren (D-MA) + Rounds (R-SD) jointly asking the administration to keep critical AI infrastructure on US soil.
Anthropic acquires Stainless (Stainless API, Inc., NY-based, founded 2022 by ex-Stripe engineer Alex Rattray) for over $300M per The Information. The startup's SDK-generation platform produced client libraries and MCP servers for OpenAI, Google, Cloudflare, Replicate, and Runway. Anthropic confirmed it will wind down all hosted Stainless products. Head of Platform Engineering Katelyn Lesse: 'Agents are only as useful as what they can connect to.' Hacker News developer pushback frames the wind-down as exclusionary; no FTC/DOJ comment yet.
Nine-member advisory jury in Oakland deliberated <2 hours. Unanimously found Musk's claims time-barred under 3-yr (charitable trust) + 2-yr (unjust enrichment) limitations. Judge Gonzalez Rogers immediately adopted as the court's finding. Musk's own 2019 testimony — 'If you've got a capped-profit situation, it hasn't violated the nonprofit's goal' — was the kill shot. Trial-record admissions: xAI/Grok trained by distilling OpenAI models; Musk tried to poach Altman to xAI. OpenAI counsel Savitt: 'weapon of a competitor who can't compete in the marketplace.' Musk counsel Toberoff: 'appeal.'
50 empty Waymos circle Battleview Drive cul-de-sacs at 6 a.m.; a Step2Kid neon flag-figure sign gridlocks 8 of them. Atlanta Mayor Andre Dickens (D) has no authority — GA Code preempts the city, the Trump-Duffy NHTSA AV Framework preempts the patchwork, and Dickens is partnering with Waymo on a pothole pilot anyway.
May 17, 2026 — China and the US together account for ~80% of global data-center electricity growth to 2030, per the IEA Energy & AI Report. The US generated a record 4,430 TWh in 2025 (+2.8% YoY) — and is projected to grow generation +50% by 2050. China is doing roughly the same buildout in five years. 10,300 US projects sit stuck in interconnection queues totaling 1,400 GW of generation + 890 GW of storage (LBNL, end-2024). FERC Order 1920 mandates 20-year regional transmission planning; DOE invoked rare authority directing FERC to issue a large-load interconnection final rule by April 30, 2026. The AI race is being lost in the substation, not the data center.
Document-preservation demand issued. $19 billion in market cap erased in a day.
Three weeks of testimony. Musk in China for closing arguments. Altman's 90-percent claim. Nadella's Microsoft testimony.
Bloomberg, May 15, 2026: AI-assisted hacking — LLM phishing, deepfaked voice/video, automated smart-contract vulnerability scanning — pushes the $130B DeFi sector 'to the brink.' April 2026: $651M in single-month theft, including Kelp DAO ($293M, via LayerZero) and Drift Protocol ($280M). North Korea's Lazarus Group: 76% of 2026 crypto-service compromises. $9B TVL drawdown on the largest DeFi lender post-hack.
Bloomberg reports Anthropic is seeking $30B+ at a $900B post-money valuation — 137% above February's Series G. ARR is $30–40B. No term sheet signed. The round would make Anthropic the most valuable private company in history.
The ACCESS model — Advancing Chronic Care with Effective, Scalable Solutions — launches July 5, 2026, with 150+ approved participants and outcome-aligned payments for AI chronic care agents. CMS is a $900B price-setter. When Medicare pays for AI, private insurers follow.
404 Media reported May 12 that a senior ICE official at the Border Security Expo in Phoenix described a Palantir-built mobile database covering ~20M people, deployed to agent iPhones. Built under the $30M sole-source ImmigrationOS award April 17, 2025 plus a $29.9M renewal Sept 25, 2025. The ELITE app pulls HHS/Medicaid records and 30+ federal datasets to return target dossiers with confidence scores. Reps. Goldman (D-NY), Wyden (D-OR), and Velázquez (D-NY) opened an 11-question congressional inquiry in April 2026.
GM began notifying 500–600 salaried IT employees Monday May 11 — primarily Warren MI and Austin TX, ~10% of GM's IT department. AI was the top-cited reason for U.S. layoffs in April 2026 (21,490 attributed cuts). The most blue-collar brand in American manufacturing just did what Amazon, Meta, and Google have been doing.
Cerebras (CBRS) raised its IPO range to $150–$160 on May 11 — book oversubscribed 20x, gross proceeds ~$4.8B, fully-diluted valuation ~$48.8B. Biden Treasury blocked the deal 18 months over G42 (UAE); cleared in October 2025 after G42 went non-voting. The structural concentration that worried CFIUS is still 86% of revenue.
Google's Threat Intelligence Group confirmed with 'high confidence' that a nation-state actor used an AI system functionally similar to Mythos to discover and exploit a zero-day 2FA vulnerability. Mass exploitation was planned — and thwarted. Mandiant's John Hultquist: 'AI-assisted offensive tooling is here.'
Santa Clara County DA Jeff Rosen filed suit alleging Meta knowingly profits from $7 billion per year in 'high-risk' scam ads — causing $2.5 billion in California consumer losses in 2024 alone, $800 million to seniors. The lawsuit challenges Section 230's shield against advertising revenue claims.
Lumma Stealer malware stole an OAuth token from Context.ai — granting read access to the company's Vercel environment variables, including production AI API keys. One stolen token. Thousands of downstream users exposed. Vercel wasn't breached — but the supply chain was. A playbook every AI startup needs to read.
EO 14365 (December 11, 2025) imposed the broadest federal AI preemption in US history — blocking state and local AI regulation for five years. Colorado AG Phil Weiser filed suit in April 2026; a federal judge granted a stay April 28. Senate passed a related moratorium 99-1. $42.45B in BEAD broadband funding is now conditioned on zero AI restrictions.
The OSTP released a 27-recommendation AI policy framework on March 20, 2026. It explicitly rejects the EU AI Act's risk-classification model in favor of voluntary, sector-specific guidance. David Sacks leads the White House AI & Crypto Council. Michael Kratsios co-chairs PCAST. Covers compute, export controls, standards, talent, and national security.
On May 4, 2026 Anthropic announced a $1.5B joint venture with Blackstone, Hellman & Friedman, and Goldman Sachs to deploy Claude across mid-market companies — embedding Anthropic Applied AI engineers inside customer engineering teams. Twenty-four hours later, Dario Amodei and JPMorgan's Jamie Dimon shared a stage in New York to launch ten financial-services agents, full Microsoft 365 integration, and a Moody's data partnership covering 600 million companies. OpenAI announced its own $10B 'Deployment Company' the same day. The AI value layer just moved.
Dr. Matthew Powell-Palm and the J. Mike Walker Department of Mechanical Engineering at Texas A&M published a Scientific Reports paper showing that tuning the glass-transition temperature of vitrification cryoprotectant solutions stops the structural cracking that has, for sixty years, made it impossible to bank a donor organ for more than a few hours. The path to a kidney bank just got measurably shorter.
On April 30 the USGS published its first quantitative assessment of lithium pegmatites in Maine and New Hampshire — about 900,000 metric tons of recoverable Li₂O. Same week: the McDermitt Caldera on the Nevada-Oregon border, holding an estimated 20-40 million metric tons, ran past 1,000 construction workers at Thacker Pass. A 44× scale gap. A clay-vs-pegmatite geology gap. A 10-year permitting gap. And behind both of them, the China supply chain — 60% of global lithium refining, 98% of LFP cathodes — that the United States is racing to break.
AWS and Cerebras Systems announced a cross-vendor disaggregated inference architecture inside Amazon Bedrock — the first hyperscale deployment to put two competing custom AI chips on the same model and route the workload to whichever one is faster at each step. Trainium handles prefill. Cerebras CS-3 — built around the wafer-scale WSE, the largest chip ever manufactured — handles decode. AWS: 'order of magnitude faster.' Launching H2 2026.
European Central Bank Governing Council member and Bank of Spain Governor José Luis Escrivá told Bloomberg in Tarragona on May 9, 2026: 'Recent developments in artificial intelligence force us to reassess the robustness of our financial infrastructure and our cybersecurity.' The ECB's mandate covers monetary policy, banking supervision (SSM), and systemic risk (ESRB). Concentration risk: 3 firms control 88% of enterprise LLM capacity. DORA designated 19 critical ICT providers. IMF's May 7 report documented AI-driven cyberattacks accelerating against financial institutions.
Claude Mythos Preview can find and exploit zero-days in every major OS and browser — 181× more effective than its predecessor. Anthropic locked it behind a vetted partner program. A Discord group got in by guessing the URL. Then GPT-5.5 shipped publicly — and on the U.K. government's cyber eval, it scored higher.