AI & The Machine — Four Fingers News

AS OF: Tue Jun 2 · 10:00 PT

Lab Scorecard — Domestic

FRONTIER LABS · FLAGSHIP MODELS · CAPITAL · HEADCOUNT · SAFETY POSTURE

LAB	FLAGSHIP	VALUATION / CAPEX	HEADCOUNT	SAFETY POSTURE
USAnthropic	Claude Opus 4.8 (May 28) · Confidential S-1 filed Jun 1 — $47B revenue run-rate, up from $9B end-2025 · Series H closed — $65B at $965B (May 28) · Mythos-class general release flagged “coming weeks” · Q2 tracking first profit — $10.9B revenue	$965B post-money (Series H, May 28) — tops OpenAI’s $852B · +$40B Google framework · $1.5B PE JV · SpaceX Colossus compute to 2029	~3,000–5,000	Runs a public Responsible Scaling Policy and keeps its Mythos-class vulnerability-discovery models gated — Anthropic says no lab can yet release one safely. Signed the CAISI pre-deployment testing MOU; its own research agenda still grants a non-zero chance the technology “kills everyone,” and the Pentagon’s IL6/IL7 list excludes the lab.
USOpenAI	GPT-5.5 (Apr 23) · GPT-5.5 Instant refresh + Codex Computer Use on Windows (May 30) · Rosalind Biodefense access tier (May 29) · Frontier Governance Framework (May 29) · Confidential IPO S-1 filed (May 22) · Reasoning model disproves Erdős conjecture (May 20)	$852B valuation (Mar) · $122B raised · $10B Deployment Company JV (May 4)	~7,700	Published a Frontier Governance Framework mapping its Preparedness Framework onto California’s frontier-AI statute and the EU Code of Practice — cyber, CBRN, manipulation and loss of control. UK AISI’s cyber eval surfaced a universal jailbreak across its malicious-query set; OpenAI patched the named findings, but a configuration issue blocked verification of the final fix.
USGoogle DeepMind	Gemini 3.5 Flash (May 19) · Gemini Spark 24/7 agent live for US AI Ultra (May 29) · Gemini 3.5 Pro due this month · Omni Flash + Antigravity + Managed Agents API (May 19) · AI Ultra repriced, new $100 tier (May 19)	$180–190B capex (Alphabet, raised Apr 29)	~6,000–8,200	Operates under Google’s Frontier Safety Framework and signed a CAISI MOU May 5 putting Gemini models under Commerce pre-deployment review, safeguards stripped for testing. Pushes SynthID provenance watermarks across its image and video generation.
USxAI	Grok 4.3 Beta (Apr 17) · Grok Build 0.1 public-beta coding model + Web Connectors (May 29) · Custom Skills (May 26) · Speech / Imagine API (May) · Grok 5 training wrapped late Apr	$20B round Jan 2026 ($230B) · SpaceX merger Feb ($1.25T combined)	~3,000 (Memphis)	No published risk or safety framework. Signed its first external-evaluation deal — a CAISI MOU May 5 — putting Grok under Commerce pre-deployment review with safeguards reduced for the probe, even as xAI’s federal challenge to state AI rules continues.
USMeta AI	Llama 4 Scout/Maverick (Apr) · Muse Spark (proprietary, Apr 8) · Meta One Plus / Premium AI chatbot subscriptions launch (May 28)	$125–145B capex 2026 (raised Apr 29)	~76,000 (-8,000 / 10%)	No formal safety framework disclosed for its open-weight Llama line. Leaked April all-hands audio has Zuckerberg defending a Model Capability Initiative that trained models on employee activity across Gmail, Chat and VS Code — said to be anonymized — which drew internal protests.

Anthropic (Claude Opus 4.8) Axios (Opus 4.8 + Mythos teaser) 9to5Google (Gemini Spark live) TechCrunch (Gemini Spark debut) OpenAI release notes (GPT-5.5 Instant + Codex on Windows) Basenor (Grok Build 0.1 + Connectors) Simon Willison (Opus 4.8 review) 9to5Mac (Opus 4.8 benchmarks) Help Net Security (Mythos-class plans) TechCrunch (Anthropic Series H $65B) Bloomberg (Series H $965B, eclipses OpenAI) Al Jazeera (Anthropic $965B) The Hill (Anthropic overtakes OpenAI) OpenAI (Frontier Governance Framework) StartupHub (Frontier Governance rollout) Techerati (OpenAI Frontier) OpenAI (Rosalind Biodefense) Axios (OpenAI biodefense program) OpenAI (Third-party evals playbook) Bloomberg (Anthropic founders into 500 richest) Bloomberg (Meta AI subscriptions) TechCrunch (Meta One Plus / Premium) Anthropic (Claude Platform on AWS) AWS (Claude Platform GA) TechCrunch (evil-AI portrayals) TechStartups (FT: Anthropic $1T) Bloomberg (Anthropic $30B/$900B, May 12) The Information (Anthropic ↔ Stainless) OpenAI (Realtime trio + GA) Google (Gemini 3.1 Flash-Lite GA) Bloomberg (Hassett FDA-style EO) Anthropic (SpaceX compute, higher limits) xAI (Compute partnership w/ Anthropic) Bloomberg (Anthropic ↔ SpaceX) SiliconANGLE (Dreaming) The New Stack (Dreaming + Outcomes) Simon Willison (Live blog) CoinDesk (June IPO flag) DefenseScoop (Pentagon-8 + Oracle) UK AISI (GPT-5.5 cyber eval) NIST CAISI (Google/MSFT/xAI MOUs) Bloomberg/WSJ (White House vs Mythos expansion) TechCrunch ($50B/$900B) Bloomberg (Anthropic $900B) CNBC (Google +$40B Anthropic) Amazon (AWS ↔ OpenAI) OpenAI (GPT-5.5) Bloomberg (OpenAI Deployment Co.) Blackstone (Anthropic JV) Anthropic (Enterprise AI services) Anthropic (Finance agents, May 5) Fortune (Anthropic Wall Street, May 5) Fortune (Anthropic vs consultants) Axios (NSA on Mythos) Fortune (Q1 capex) TechCrunch (Anthropic tops Ramp AI Index) Axios (Anthropic overtakes OpenAI adoption) Ramp (AI Index, May 2026) Inc. (Claude for Small Business) The Register (Claude for Small Business) The AI Insider (Anthropic $950B framing) Anthropic (Gates Foundation $200M) Gates Foundation (Anthropic partnership) Anthropic (Project Glasswing) CyberScoop (OpenAI Daybreak) The Hacker News (Daybreak / GPT-5.5-Cyber) Anthropic (PwC alliance expanded) SiliconANGLE (PwC 30K trained) InfoWorld (Anthropic Agent SDK meter) The New Stack (Agent SDK credits split) SiliconANGLE (programmatic credit pool) LawSites (Claude for Legal) TechCrunch (Anthropic legal push) TechCrunch (ChatGPT Personal Finance + Plaid) OpenAI (Personal Finance) Bloomberg (OpenAI ↔ Plaid) OpenAI (Codex on mobile) TechCrunch (Codex on mobile) Basenor (Grok Build beta) Techloy (Grok Build vs Claude Code) OpenAI (Deployment Company) Bloomberg (OpenAI ↔ Tomoro) CNBC (Disruptor 50: Anthropic No. 1) CNBC (Disruptor 50 full list) Gizmodo (Mythos confidentiality loosened) Reuters/KFGO (Mythos sharing) Techzine (Mythos partner sharing) Anthropic release notes (SEC + cache diagnostics, May 19) Anthropic (Code with Claude London) Anthropic (KPMG global alliance, 276K) Anthropic (Stainless acquisition) InfoWorld (Stainless close) CNBC (Gemini 3.5 / Spark / Omni) Google (I/O 2026 developer keynote) CNBC (Anthropic Q2 $10.9B / first profit) Axios (Anthropic pays SpaceX $15B/yr) Reuters/Yahoo (Anthropic profit, $1.25B/mo) OpenAI (model disproves geometry conjecture) TechCrunch (OpenAI Erdős proof) Google (Gemini 3.5 pricing) Dataconomy (AI Ultra cut to $200, new $100 tier) PR Newswire (Fujitsu ↔ Anthropic + OpenAI) Anthropic (Glasswing initial update) Help Net Security (Glasswing 10K+ findings) CSO Online (Glasswing 10K vulns) SiliconANGLE (Claude Security public beta) Anthropic (KiYoung Choi, Korea Director) Korea Herald (Anthropic Seoul office) OpenAI (Election safeguards 2026) Axios (OpenAI election defenses) FT/Investing (Anthropic agrees $30B/$900B) Anthropic (confidential draft S-1) TechCrunch (Anthropic files to go public)

China & International Watch

FRONTIER LABS · CHIP ACCESS · BENCHMARKS · GOVERNMENT SUPPORT

LAB	FLAGSHIP	CHIP ACCESS	BENCHMARK	NOTES
CNDeepSeek	V4 Preview (Apr 24) · Pro 1.6T (49B active) + Flash 284B (13B active), 1M ctx, MIT	Huawei Ascend 950 supernode clusters — part of V4 training done on Ascend	#1 open-weight on Vals AI Vibe Code · 80.6% SWE-bench Verified (within 0.2pp of Opus 4.6)	Made its 75% V4-Pro cut permanent (~$0.87 output per M tokens, below GPT-5, Opus and Gemini list). Its first external round — near $7.35B at up to $50B — is in late talks, with China’s state AI fund negotiating to lead and Tencent eyeing up to 20%. CAISI rates V4 Pro the top PRC model, ~8 months off the frontier.
CNAlibaba (Qwen)	Qwen 3.7-Max (May 20) · Qwen 3.6 Plus (Apr 2)	Alibaba Cloud + in-house Zhenwu M890 (144GB HBM3) · bulk Ascend 950PR orders	Arena #13 text / #7 math (3.7-Max-Preview) · #1 OmniDocBench (91.2)	Launched Qwen 3.7-Max at its Hangzhou Cloud Summit May 20 — a frontier agentic model billed for 35-hour runs and 1,000-plus tool calls, #7 on Arena math as a preview. The same morning the Pingtouge arm unveiled the Zhenwu M890 processor (144GB HBM3), claimed at 3× the H20 — a full-stack reply to export controls.
CNBaidu (ERNIE)	ERNIE 5.1 (May 9) · params ~1/3 of 5.0, ~1/2 active	Kunlun chip (in-house) + legacy Nvidia	#1 CN · #4 global on LMArena search	ERNIE 5.1 (May 9) cuts parameters to about a third of 5.0 at a claimed 6% of comparable training cost; it tops the domestic LMArena search board, sits 4th globally, and is rolling out across 10-plus Chinese platforms. At Create 2026, Robin Li reframed the race around agents and “self-evolution.”
CNByteDance	Doubao 2.0 (Feb) · Seedance 2.0 (video)	Bulk Ascend 950PR orders	Seedance: #1 img-to-video on Artificial Analysis	China’s most-used consumer AI app — Doubao around 155M weekly actives. Seedance in CapCut; Hollywood IP complaints paused the global rollout.
EUMistral (France)	Mistral Medium 3.5 (Apr 29) · 128B dense, modified MIT, 256K ctx	Nvidia GB300 · $830M debt for Paris DC (13,800 GPUs)	Open-weight frontier · on Microsoft Foundry day-one	Medium 3.5 (Apr 29) hits 77.6% SWE-bench Verified and self-hosts on four GPUs. CEO Mensch told CNBC May 28 the firm is exploring its own chips to control more of its stack; it also bought Emmi AI for an industrial/physics-AI vertical, alongside €4B in France/Sweden data centers. Macron: €109B AI infra; 200 MW across Europe by 2027.

CNBC (Mistral exploring own chips) TFN (DeepSeek: Tencent up to 20%) SCMP (DeepSeek first round) TechBriefly (DeepSeek $7.35B / Big Fund III) TechCrunch (DeepSeek $45B step) Microsoft Foundry (DeepSeek V4) NIST CAISI (V4 Pro eval) TrendForce (V4 Day-0 multi-vendor) Tom’s Hardware (V4 + IP claims) CNBC (DeepSeek V4) LMSys (V4 Day-0 SGLang) Vals AI (V4 Pro index) InfoWorld (Baidu ERNIE 5.1) CnTechPost (ERNIE 5.1 cost) TechNode (Baidu Create 2026) TestingCatalog (Mistral Medium 3.5) SCMP (Qwen 3.7 previews on Arena) Startup Fortune (Qwen 3.7 push) Manila Times (Alibaba full-stack AI upgrade) Quartz (Zhenwu M890 + Qwen 3.7-Max) Wccftech (Zhenwu M890 specs vs H20) Business Standard (Alibaba chip + LLM) Engadget (DeepSeek 75% V4-Pro permanent) Caixin (DeepSeek V4-Pro permanent cut) InfoWorld (DeepSeek pricing war escalates)

Compute & Capex Tracker

HYPERSCALER SPEND · GPU ALLOCATIONS · POWER DEALS · SOVEREIGN FUNDS

$650B+

2026 HYPERSCALER CAPEX

Q1 prints landed Apr 29 after close: META raised to $125–145B (was 115–135), GOOG raised to $180–190B (was 175–185), MSFT FY ~$190B with Q4 alone >$40B, AMZN ~$200B; ~75% AI-tagged. Only Google closed up; META slid 9%, with the cut tied to 8,000 confirmed layoffs. Apple Q2 (Apr 30) printed +17% revenue and pointed Siri to a custom 1.2T Gemini build — the only Big Tech sitting out the GPU build. Bloomberg reporting carried May 12 puts a third to half of US data centers planned for 2026 as delayed or canceled, with grid hookups stretching up to five years — the widest gap yet between dollars committed and watts energized. Nvidia’s May 20 print is the supply-side mirror: Q1 FY27 (ended Apr 26) revenue a record $81.6B, up ~85% YoY, with data center alone $75.2B (+92% YoY, +21% sequential) on the Blackwell 300 ramp; the company raised its dividend and buyback. Dell’s May 28 print is a second supply-side mirror: record $43.8B revenue (+88% YoY), $16.1B AI-server revenue and $24.4B in AI orders booked, with FY27 AI-server guidance lifted to $60B — the stock’s best day ever (~+32%).

220K GPUs

COLOSSUS DEAL SURFACES IN SPACEX S-1

SpaceX’s May 20 IPO prospectus put hard numbers on the deal first struck May 6: Anthropic pays about $1.25B a month through May 2029 — roughly $45B total — for Colossus 1 and 2 in Memphis (220,000-plus NVIDIA GPUs, 300+ MW). Daniela Amodei said Anthropic is scaling onto NVIDIA GB200 capacity in Colossus 2 through June. The building was raised to train Grok; it now serves Claude. SpaceX’s SpaceXAI segment (the former xAI) showed $818M in Q1 revenue against a $2.47B operating loss. Either side can exit on 90 days’ notice; Claude Code rate limits keep rising as GB200 capacity comes online.

$67B

NEXTERA TO BUY DOMINION (MAY 18)

All-stock combination announced May 18 (Bloomberg, CNBC, WaPo, NBC, Fox Business) folds Dominion — the utility behind Northern Virginia “Data Center Alley,” the world’s largest data-center market — into NextEra. Dominion already has ~51 GW of contracted data-center capacity with Alphabet, Amazon, Microsoft, Meta, Equinix, CoreWeave and CyrusOne. Combined entity: ~$400B enterprise value including debt, 10M customers across Florida, Virginia, the Carolinas; #1 in renewables and battery storage, US leader in natural-gas generation, #2 in nuclear. NextEra targeting 30+ AI data-center hubs. NextEra holders take 74.5%, Dominion 25.5%; Ketchum stays CEO of the combined entity, Blue runs regulated utilities. Closing pegged 12–18 months out across FERC, DOJ and PUCs in FL/VA/NC/SC; lawmakers and consumer advocates flagged risk that residential ratepayers subsidize data-center expansion. Context: IEA still pegs 2026 global DC power at 1,100 TWh (+18% vs Dec 2025); Bloom–Oracle 2.8 GW SOFC; MSFT/Constellation $16B / 2 GW; Meta 6.6 GW across Vistra, Oklo, TerraPower. CNBC (May 18): European industrial electricity ran roughly 2× US and 50% above China and India last year (IEA); securing DC capacity in Frankfurt, London, Amsterdam, Paris and Dublin set to cost 12% more in 2026 (CBRE). OpenAI paused its UK Stargate leg partly over energy cost.

$965B

ANTHROPIC CLOSES SERIES H, PASSES OPENAI

Anthropic closed a $65B Series H May 28 at $965B post-money (TechCrunch, Bloomberg, Al Jazeera, The Hill, Business Standard) — co-led by Altimeter, Dragoneer, Greenoaks, Sequoia, Capital Group, Coatue and D1, with Baillie Gifford, Blackstone, Brookfield, D.E. Shaw Ventures, DST and Fidelity participating. The mark eclipses OpenAI’s $852B March valuation and lands as Anthropic tracks its first profitable quarter (Q2 revenue near $10.9B, op profit ~$559M) and signals an October IPO. On Jun 1 Anthropic confidentially filed a draft S-1 with the SEC, disclosing a revenue run-rate past $47B (up from $9B at end-2025). The IPO calendar tightens: SpaceX opens its roadshow Jun 4 and targets a Jun 12 Nasdaq debut (SPCX) at $1.75T — a ~$75B raise, the largest IPO on record, with 30% reserved for retail; OpenAI filed a confidential S-1 May 22 (Fortune, Axios, AI Weekly), advised by Goldman Sachs and Morgan Stanley, with a Q4 listing framed $852B–$1T — financials sealed until ~15 days before the roadshow. Whoever prices first sets the comparable. DeepSeek’s ~$7.35B Big Fund III round at $50B is still in late talks — the largest Chinese AI raise on record; Hillhouse and Tencent in.

TechCrunch (Anthropic Series H $65B) Bloomberg (Series H $965B) Business Standard (Anthropic $965B) CNBC (Dell +32%, AI servers) Bloomberg (Dell $60B AI-server guide) Shacknews (Series H $65B / $965B) SCMP (DeepSeek $50B / Big Fund III) AWS (Claude Platform GA) TechStartups (FT: Anthropic $1T) Anthropic (SpaceX compute, higher limits) xAI (Compute partnership w/ Anthropic) The Decoder (Colossus 1 / 220K GPUs) Engadget (Claude Code limits doubled) CoinDesk (June IPO flag) Fortune (Q1 capex prints) Yahoo ($650B Q1) StockTitan (Nvidia Q1 FY27: $81.6B / $75.2B DC) IEA 1,100 TWh CNBC (NextEra to buy Dominion, $67B) Bloomberg (NextEra–Dominion) WaPo (Dominion–NextEra merger) CNBC (Europe AI energy gap) Bloom/Oracle 2.8 GW Bloomberg (OpenAI Deployment Co. $10B) Blackstone (Anthropic $1.5B JV) Stanford HAI Tech Insider (US data center delays) The AI Insider (Anthropic $950B) CNBC (Anthropic Q2 $10.9B / first profit) Axios (Anthropic ↔ SpaceX $1.25B/mo) Morningstar (SpaceX S-1 filing) CNBC (SpaceX IPO: roadshow Jun 4, price Jun 11, list Jun 12) CNBC (SpaceX 30% retail allocation) CNBC (OpenAI confidential IPO) CNBC (Anthropic S-1, $47B run-rate) NPR (Anthropic IPO filing)

Model Releases

LAST 30 DAYS · 255 MODELS SHIPPED Q1 2026

JUN 1FUNDAnthropic confidentially files its IPO paperwork — Anthropic submitted a draft S-1 to the SEC on Jun 1 (Anthropic, TechCrunch, CNBC, CNN, NBC, NPR), starting the clock on a listing it has framed for the fall. The confidential route keeps share count and price open and seals financials until roughly 15 days before a roadshow, but the draft carries one fresh number: a revenue run-rate past $47B, up from about $9B at the end of 2025, set against the $965B Series H closed May 28. OpenAI is on the same trillion-dollar IPO track; SpaceX floats first, opening its roadshow Jun 4.
MAY 30NEWOpenAI refreshes GPT-5.5 Instant and puts Codex on Windows — the May 30 update gives GPT-5.5 Instant clearer, less bullet-heavy answers and moves writing and code into inline blocks, retiring Canvas in the Instant and Thinking tiers (OpenAI release notes). Codex gains Computer Use on Windows — it can see, click and type in Windows apps — plus remote continuation from mobile or Mac and new Codex Profiles for usage and token tracking. Product polish, not a new model, a week after the confidential S-1.
MAY 29NEWGoogle’s Gemini Spark agent goes live — ten days after its I/O debut, Spark opened to U.S. Google AI Ultra subscribers (9to5Google, TechCrunch). Where the chatbot stops when the tab closes, Spark runs on dedicated Google Cloud VMs around the clock and acts across connected apps with the machine off; it ships with MCP links to Canva, OpenTable and Instacart, with Adobe, Spotify, GitHub, Notion and Slack slated for summer. Google tags it Beta.
MAY 29NEWxAI ships Grok Build 0.1 and Web Connectors — Grok Build 0.1, pitched as a fast coding model, entered public beta through the xAI API (Basenor, xAI docs); the same push wired Grok Web into SharePoint, Outlook, OneDrive, Google Workspace, Notion, GitHub and Linear. The pair lands three days after Custom Skills, filling out a consumer-agent stack against Claude Code and Codex.
MAY 29NEWOpenAI launches the Rosalind Biodefense Program — the lab turns GPT-Rosalind, its April life-sciences reasoning model, into a sponsored institutional tier: OpenAI covers API costs for vetted partners building pandemic-preparedness tools (Axios, OpenAI, RDWorld, AI Weekly). Early partners include Lawrence Livermore National Laboratory, Johns Hopkins Applied Physics Laboratory, CEPI, Fourth Eon and SecureDNA; CEPI is applying it to its 100 Days Mission for faster vaccine development, including against a current Bundibugyo Ebola outbreak. OpenAI says it briefed the White House and federal agencies before launch.
MAY 28NEWAnthropic ships Claude Opus 4.8 alongside the Series H close — 41 days after Opus 4.7. Agentic coding rises from 64.3% to 69.2%; multidisciplinary reasoning with tools climbs from 54.7% to 57.9%; agentic computer use 82.8% to 83.4%; the knowledge-work score 1753 to 1890 (Anthropic, Axios, 9to5Mac, Simon Willison). Pricing holds at Opus 4.7 levels; Fast mode for Opus 4.8 lands at roughly a third the cost of prior fast modes; Dynamic Workflows in Claude Code enter research preview. Anthropic calls it “a modest but tangible improvement” with sharper judgement and more honesty about its progress, and flags Mythos-class vulnerability-discovery models for a general release in “the coming weeks” once safeguards land.
MAY 28FUNDAnthropic closes Series H at $65B / $965B post-money — the round eclipses OpenAI’s $852B March mark to make Anthropic the most valuable private AI company (TechCrunch, Bloomberg, Al Jazeera, The Hill, Business Standard, Shacknews). Co-leads: Altimeter, Dragoneer, Greenoaks, Sequoia, Capital Group, Coatue, D1. Institutional participation: Baillie Gifford, Blackstone, Brookfield, D.E. Shaw Ventures, DST Global, Fidelity. Anthropic says proceeds go to “safety and interpretability research, compute to meet growing demand for Claude, and the products and partnerships our customers rely on.” The round closes a path that ran $30B→$45B→$50B→$65B across the month and likely lands as the last private raise before an October IPO window.
MAY 28NEWMeta puts a price tag on Meta AI — Meta One Plus at $7.99/mo for heavy image, video and extended-reasoning users; Meta One Premium at $19.99/mo for higher volumes, rolling out first in Singapore, Guatemala and Bolivia with more markets to follow (Bloomberg, TechCrunch, Yahoo, The Star, Breitbart). Free Meta AI continues with hard caps. The launch sits alongside Facebook Plus / Instagram Plus / WhatsApp Plus tiers ($2.99–$3.99/mo) and is the first consumer revenue line aimed squarely at offsetting the $125–145B 2026 capex bill.
MAY 27NEWFujitsu signs both Anthropic and OpenAI for Japan’s enterprise build — the IT giant struck a strategic partnership with Anthropic, pairing Claude with its mission-critical systems work, widening its Forward Deployed Engineer model and putting Claude in front of every Fujitsu Group employee; a parallel OpenAI collaboration folds GPT into the same service portfolio. Both deals pitch the safety and reliability of social infrastructure — a Japanese house hedging across both frontier labs at once.
MAY 22FUNDOpenAI files its IPO paperwork confidentially — OpenAI lodged a confidential S-1 with the SEC on May 22, advised by Goldman Sachs and Morgan Stanley, with a Q4 2026 listing — possibly September — at a valuation framed between its $852B private mark and $1T (Fortune, Axios, AI Weekly). The confidential route starts the SEC clock without disclosing financials, which stay sealed until roughly 15 days before the roadshow. Anthropic, tracking its first profit, has signaled October; SpaceX, public since its May 20 filing, now opens its roadshow Jun 4 toward a Jun 12 Nasdaq float. Three frontier-adjacent debuts now share a six-month window.

Policy Watch

US FEDERAL · STATE · EU · CHINA

JUN 2NEWMicrosoft makes Copilot multi-model at Build — Satya Nadella opened Build 2026 (Jun 2–3, Fort Mason) casting agents as “the new operating system for work” (Microsoft, Windows News). Copilot’s rebuilt orchestration layer now routes across OpenAI, Anthropic’s Claude and on-device small models, tenant-configurable, ending the OpenAI exclusivity that had held since 2023; Copilot Studio and Excel’s Agent Mode gain Claude as an option. Microsoft also shipped its own MAI model family, including a coding model for GitHub Copilot, plus Agent Mode across Word, Excel, Teams and Outlook and a .NET/Python Agent Framework on the A2A protocol.
MAY 29USOpenAI publishes a third-party evaluation playbook — the same day as its Frontier Governance Framework, OpenAI puts out guidance on what makes independent evaluations of frontier models valid, arguing the “harness” — the tools, state and retry scaffolding wrapped around a model — can swing measured performance enough to decide whether a capability shows up at all, and calling for standardized reporting on harness choices and hazard mitigations (OpenAI, StartupHub). It is the third governance move of OpenAI’s day, alongside the Frontier Governance Framework and the Rosalind Biodefense access tier — a dense safety paper trail a week after the confidential S-1.
MAY 29USOpenAI publishes a Frontier Governance Framework — OpenAI carries its internal Preparedness Framework into a public governance document that maps practice to California’s Transparency in Frontier AI Act and the EU AI Act Code of Practice for General Purpose AI (OpenAI, StartupHub, Techerati). Coverage spans cyber offense, CBRN, harmful manipulation and loss of control, plus model reporting, security risk management, incident response, external expert input and update cadence. The framework lands a day after Anthropic’s Series H close and one week after OpenAI’s own confidential S-1 filing, lining the regulatory paper trail up with the IPO calendar.
MAY 27USOpenAI lays out 2026 election guardrails — the company will surface live AP election-night results in U.S. and Brazilian ChatGPT this fall and partner with Democracy Works on voter-registration and polling-place queries, opens Codex Security and Trusted Access for Cyber to registered U.S. voting-system manufacturers, and is briefing NASS and NASED on the latest defensive tooling. SynthID watermarks come to ChatGPT image generation via Google DeepMind; OpenAI is endorsing the Protect Elections From Deceptive AI Act and the Preparing Election Administrators for AI Act. The package frames misinformation, deepfakes and provenance as enterprise-side responsibilities, not just platform-side.
MAY 27USThe labs split on the jobs question as the IPOs line up — Sam Altman told a Commonwealth Bank audience he was “delighted to be wrong,” that the entry-level white-collar wipeout he forecast hasn’t shown up and a “jobs apocalypse” now looks unlikely; Fortune reports Dario Amodei has softened the same prophecy. The same week Anthropic co-founder Chris Olah told the Vatican’s AI ethics conference there is “a real possibility that AI will displace human labor at very large scale.” Yale’s Budget Lab found no meaningful unemployment shift through March for high-exposure jobs. The doom rhetoric cools as both labs court IPO investors.
MAY 26USProject Glasswing crosses 10,000 critical-severity findings — Anthropic’s initial Glasswing update reports that Claude Mythos Preview, run by roughly 50 partner organizations, has now surfaced more than 10,000 high- or critical-severity vulnerabilities in systemically important software. Independent reviewers validated more than 90% of 1,752 sampled findings. Claude Security entered public beta for Enterprise (already used to patch 2,100-plus vulns in three weeks); Mythos itself stays gated — Anthropic says no lab has yet developed safeguards strong enough for public release of a model at that capability. The number lands the week after Verizon joined Glasswing and a month after OpenAI’s less-secretive Daybreak.
MAY 25CNDeepSeek makes its 75% V4-Pro price cut permanent — the discount, originally set to expire May 31, is locked in: API pricing runs about $0.0036 input to $0.87 output per million tokens, below GPT-5, Claude Opus and Gemini list prices. Anthropic’s own May policy paper grants Beijing is “winning in global adoption on cost”; OpenRouter traffic share for Chinese models is past 60%, up from ~1% in 2024. Big Fund III talks on DeepSeek’s ~$7.35B raise at a $50B valuation continue.
MAY 21USTrump pulls the AI oversight order hours before signing — the ceremony was on the schedule and invitations were out when Trump called it off, telling reporters he “didn’t like certain aspects of it” and worried the order “could have been a blocker” to the U.S. lead over China. The shelved draft set a voluntary framework asking frontier developers to hand covered models to the government up to 90 days before release and to give critical-infrastructure operators such as banks pre-release access, with a Treasury-led clearinghouse hunting flaws in unreleased models. Industry had pushed for 14 days against the draft’s 90. Axios reports Musk, Zuckerberg and former AI czar David Sacks each reached Trump directly between Wednesday night and Thursday morning, outside the policy process, warning the review could slow the build; the national-security staff who drafted it had no comparable line. No new date is set; the FDA-style proposal has now slipped repeatedly since May 13.
MAY 20USSupply-chain worm walks 3,800 repos out of GitHub — GitHub confirmed the TeamPCP group (UNC6780) exfiltrated about 3,800 internal repositories after a trojanized Nx Console VS Code extension sat on the Marketplace for 18 minutes May 18, harvesting GitHub and npm tokens, AWS keys, 1Password vaults and Claude Code config files. The same Mini Shai-Hulud campaign (CVE-2026-45321, CVSS 9.6) reached 170-plus npm packages, two OpenAI developer devices, a Mistral machine now under a Monero extortion demand and the European Commission’s public site; OpenAI revokes its macOS signing certificate June 12. The breach reads as the concrete infrastructure risk the shelved EO named.
MAY 14USColorado legislature repeals and replaces the AI Act — SB 26-189 cleared both chambers with bipartisan support and now sits on Gov. Polis’s desk; he told reporters May 12 he’ll sign. The replacement swaps the 2024 law’s risk-management programs, annual impact assessments and broad algorithmic-discrimination duties for a narrower notice-and-transparency framework activating in 2027. xAI’s federal-court challenge survives regardless: preliminary injunction motion due within 28 days of finalized rulemaking under the successor statute; DOJ stays in.
JUN 8UPCOMINGApple WWDC 2026 keynote — iOS 27, iPadOS 27 and macOS 27 expected to ship Siri 2.0 with a chatbot redesign, Dynamic Island integration, conversation history, and a new Extensions system piping queries into third-party assistants (Claude, Gemini and Grok flagged). Apple is reportedly leaning on Google’s custom 1.2T Gemini build under a ~$1B deal.
DEC 2EUEU AI Act Art. 50(2) watermarking (upcoming) — the May 7 omnibus deal pushed provider-side marking of AI-generated content from Aug 2 2026 to Dec 2 2026 and set the same date for the new Art. 5 prohibition on AI-generated CSAM and non-consensual intimate imagery.

Bird & Bird (EU Omnibus May 7 deal) Bloomberg (Anthropic $30B/$900B, May 12) The Information (Anthropic ↔ Stainless) Google (Android Show: I/O Edition 2026) Google (Gemini Intelligence) TechCrunch (Android Show recap) CNBC (Gemini vs Apple) Isomorphic Labs ($2.1B Series B) FierceBiotech (Isomorphic Series B) Axios (WH AI EO stalled) Axios (Trump EO: early gov access) CNN (Trump could sign Thursday) CNBC (Trump postpones AI order) WaPo (Trump calls off AI order) Axios (why the order was pulled) IAPP (Omnibus next steps) Anthropic (Claude Platform on AWS) AWS (Claude Platform GA) TechCrunch (Anthropic evil-AI fix) Nature (Elsevier joins Llama class) SCMP (DeepSeek $50B / Big Fund III) xAI (May 15 model retirement) TechStartups (FT: Anthropic $1T) OpenAI (Realtime trio + GA) Google (Gemini 3.1 Flash-Lite GA) Zyphra (ZAYA1-8B on AMD) Bloomberg (Hassett FDA-style EO) The Hill (Hassett “like an FDA drug”) The Hill (Wiles walks back) Microsoft Foundry (DeepSeek V4) Anthropic (SpaceX compute, higher limits) xAI (Compute partnership w/ Anthropic) Bloomberg (Anthropic ↔ SpaceX) SiliconANGLE (Dreaming) The New Stack (Dreaming + Outcomes) Engadget (Claude Code limits doubled) Simon Willison (Live blog: Code w/ Claude) CoinDesk (June IPO flag) Anthropic (Finance agents) Fortune (Anthropic Wall Street stack) Bloomberg (Anthropic finance agents) Axios (Amodei + Dimon) NYT/USNews (WH pre-release vetting) NIST (CAISI MOUs Google/MSFT/xAI) CNN (Government early access) Axios (Frontier testing pivot) Anthropic (Code with Claude SF) Bloomberg (WH AI working group) Bloomberg (OpenAI Deployment Co.) Blackstone (Anthropic $1.5B JV) Anthropic (Enterprise AI services) Fortune (Anthropic JV vs consultants) Yahoo (OpenAI/Anthropic separate PE JVs) Axios (NSA on Mythos) Bloomberg/WSJ (Mythos opposition) TechCrunch (Anthropic $50B/$900B) VentureBeat (Poolside Laguna XS.2) NVIDIA (Nemotron 3 Nano Omni) Amazon (OpenAI on AWS) Modulos (EU Omnibus trilogue failed) Implicator (12-hour collapse) WUSF (FL AG widens probe) CNBC (Google ↔ Anthropic $40B) TrendForce (V4 Day-0) TestingCatalog (Mistral Medium 3.5) TestingCatalog (Jupiter v1-p / Code with Claude) Microsoft (Agent 365 GA) Qwen (Fireworks partnership) CNN (Pentagon-8 / Anthropic excluded) DefenseScoop (Pentagon-8 + Oracle) CNBC (CTO Michael on Anthropic) UK AISI (GPT-5.5 cyber eval) NIST CAISI (DeepSeek V4 Pro eval) OpenAI (Trusted Access for Cyber) TechCrunch (Anthropic 48-hour allocations) TechCrunch (Anthropic tops Ramp AI Index) Inc. (Claude for Small Business) InfoWorld (Baidu ERNIE 5.1) TechNode (Baidu Create 2026) Anthropic (Gates Foundation $200M) PYMNTS (Anthropic ↔ Gates Foundation) CyberScoop (OpenAI Daybreak) The Hacker News (OpenAI Daybreak) Let’s Data Science (Verizon joins Glasswing) Consilium (EU omnibus provisional deal) OpenAI (Personal Finance) Bloomberg (OpenAI ↔ Plaid) Decrypt (Kimi WebBridge) OSFY (Kimi WebBridge) Anthropic (PwC alliance expanded) PwC (Anthropic alliance) The Register (Agent SDK billing) VentureBeat (Anthropic reinstates OpenClaw — with a catch) OpenAI (Codex on mobile) Basenor (Grok Build beta) LawSites (Claude for Legal) OpenAI (Deployment Company + Tomoro) Bloomberg (OpenAI ↔ Tomoro) eWeek (Google I/O 2026) Nixon Peabody (CO SB 189 passes) CPR (Polis to sign CO SB 189) Norton Rose Fulbright (xAI v. CO + DOJ) Tom’s Guide (WWDC 2026 preview) Marketing Trending (iOS 27 AI extensions) WaPo (Pope ↔ Anthropic encyclical) NCR (Pope Leo AI encyclical) OSV (Magnifica Humanitas) CNBC (NextEra ↔ Dominion $67B) Bloomberg (NextEra ↔ Dominion) CNBC (Europe AI energy gap) Techmeme/Forbes (Babuschkin River AI $1B) LDS (River AI) Anthropic (Code with Claude London) CNBC (Disruptor 50 list) CNBC (Anthropic No. 1) Gizmodo (Mythos confidentiality) Reuters/KFGO (Mythos sharing) SiliconANGLE (Dust $40M Series B) Tech.eu (Dust multiplayer AI) Android Authority (I/O 2026 preview) AIxploria (Gemini 4.0 / Omni) SCMP (Qwen 3.7 previews) CNBC (Nvidia H200 cleared to China) TIME (Trump-Xi AI summit) National CIO Review (NextEra-Dominion grid) Microsoft (Build 2026) CNBC (OpenAI confidential IPO filing) Fortune (OpenAI files confidential S-1, May 22) StockTitan (Nvidia Q1 FY27 record) TechCrunch (Anthropic $1.25B/mo to SpaceX) Morningstar (SpaceX S-1: spending, losses) TechCrunch (Karpathy joins Anthropic) CNBC (cheap AI vs IPO valuations) The Hacker News (GitHub repos breached via Nx Console) VentureBeat (TeamPCP supply-chain worm) Time (Jack Clark Oxford lecture) Axios (Anthropic Institute intelligence-explosion agenda) 9to5Google (CapCut ↔ Gemini) eWeek (Gemini + Adobe/Canva/CapCut) eWeek (Meta employee tracking before layoffs) TechStory (Zuckerberg MCI leaked audio) Fortune (Altman & Amodei walk back jobs apocalypse) Axios (OpenAI vs Anthropic on AI jobs) Time (Altman: jobs apocalypse unlikely) PR Newswire (Fujitsu ↔ Anthropic + OpenAI) FT/Investing (Anthropic agrees $30B/$900B terms) OpenAI (Election safeguards 2026) Axios (OpenAI election defenses) CyberScoop (OpenAI election cyber) Help Net Security (Glasswing 10K+ findings) Anthropic (Glasswing initial update) Anthropic (Korea Representative Director) Engadget (DeepSeek 75% permanent cut) Basenor (xAI Grok Custom Skills, May) Anthropic (Claude Opus 4.8) Axios (Opus 4.8 + Mythos teaser) Gizmodo (Opus 4.8 + Mythos-class) Help Net Security (Mythos-class general release) TechCrunch (Anthropic Series H $65B) Bloomberg (Series H $965B) Al Jazeera (Anthropic $965B) The Hill (Anthropic overtakes OpenAI) OpenAI (Frontier Governance Framework) StartupHub (Frontier Governance) Techerati (Frontier signals enterprise shift) OpenAI (Rosalind Biodefense) Axios (OpenAI biodefense program) OpenAI (Third-party evals playbook) RDWorld (Rosalind early access) Bloomberg (Meta AI subscriptions) TechCrunch (Meta One Plus / Premium) The Star (Meta AI subscriptions) Microsoft (Anthropic joins Copilot Studio) Windows News (Build 2026 agents)

SAY vs. DO — The AI Race This Week

SAY: The frontier talks agents and guardrails. Microsoft opens Build casting agents as “the new operating system for work,” wires Claude into Copilot beside OpenAI and ships its own MAI models; Anthropic still flags a Mythos-class release “in the coming weeks”; Google holds Gemini 3.5 Pro for this month. DO: The money races for the exits. Anthropic confidentially files its S-1 (Jun 1), disclosing a revenue run-rate past $47B — up from $9B at end-2025 — on its $965B mark; OpenAI’s confidential S-1 has stood since May 22; SpaceX opens its roadshow Jun 4 toward a Jun 12 float near $1.75T. WATCH: which lab prices first; whether SpaceX’s Jun 11 pricing holds; Mythos timing; the shelved White House oversight order.

← Back to the front page