Top Stories
Trump scraps AI safety executive order after billionaire phone calls. A signing ceremony was already scheduled when 11th-hour calls from David Sacks, Elon Musk, and Mark Zuckerberg convinced Trump to pull the order. Sacks argued it would make the US less competitive against China. The EO was actually voluntary — "nowhere does it say mandatory" — but tech leaders lobbied against even that. Senator Mark Kelly: "America cannot lead in AI if our policy is determined by whichever billionaire gets the president on the phone last." White House insiders say Mythos specifically "freaked out" senior officials including Secretary Bessent, given its offensive cyber capability. Pentagon's Hegseth pushing for stricter regulation. 152K views on MS NOW alone. (MS NOW — "Trump faces BACKLASH after scrapping AI executive Order"; multiple sources)
Grok V9 finishes training — 1.5T parameters, fed Cursor programming data. Musk announced May 24 that Grok V9 (3x current model size) completed training, with public release in 2-3 weeks. The bombshell: xAI fed massive amounts of Cursor IDE interaction data — prompts, code edits, debugging sessions, multi-file collaboration patterns from 67% of Fortune 500 companies. SpaceX has a $60B option to acquire Cursor (with a $10B cooperation fee if they don't exercise it). Grok Build (terminal agent, 8 parallel sub-agents, $300/mo) launched May 14. V8 Small (500B) going open source by year end. But Grok still trails badly: 6% enterprise adoption vs OpenAI 55%, Anthropic 47%. SpaceX NASDAQ IPO June 12 at $1.75T target — the V9 release timing is deliberate. (AI Revolution — "Elon Musk Just Shocked OpenAI With Grok 5")
Two Nature papers: AI autonomously discovers new disease treatments. Google's Co-Scientist — a multi-agent system with generation, reflection, proximity, evolution, and ELO-ranked tournament agents — independently discovered drug repurposing candidates for acute myeloid leukemia. Standout: CUR-6, a drug with zero prior cancer links, was 18x more effective at killing leukemia stem cells than normal cells. AI also found synergistic 3-drug combinations that no human had tested. Both papers validated with real lab experiments. This is genuine autonomous scientific discovery, not just literature search. (AI Search — "The acceleration is here!")
June model collision: 4-5 frontier releases in one month. Claude Sonnet/Opus 4.8 IDs on Vertex AI. GPT 5.6 leaks (internal tags: Iris Alpha, Ember Alpha, Beacon Alpha) with 85% Polymarket probability for June. Gemini 3.5 Pro + new X-High thinking variant. MiniMax M3 with sparse attention. All while DeepSeek V4 Pro's permanent pricing ($0.44/$0.87 per 1M tokens) undercuts everyone by 10-30x. Developing story from May 25-27 — now confirmed from multiple independent sources. (WorldofAI — "Gemini 3.5 Pro X-High, MiniMax M3, DeepSwe..."; Your AI Guy — "June 2026 AI: Claude 4.8, GPT-5.6, Gemini 3.5 Pro All Landing at Once")
"AI job apocalypse is cancelled" — Silicon Valley reverses its own prediction. Both Altman (Bank of Australia conference) and Amodei have dramatically softened their job loss messaging. Altman: "I don't think we're going to have the kind of jobs apocalypse that some companies in our space talk about." Amodei: "If you automate 90% of the job, the 10% expands and now that's a new full job." Jevons paradox in action — Dan Shipper at Every reports more work than ever despite massive automation. Wes Roth and Fortune both note the timing: both companies approaching IPOs. (Wes Roth — "The 'AI Job Apocalypse' is CANCELLED!")
Model Updates
- Qwen 3.7 Max debuts at #4 on Code Arena — first Chinese model in global coding top 5. Beats GPT 5.5 and Gemini 3.5 Flash. Designed as an agent foundation model: sustained 35-hour autonomous coding sessions, 1,158 tool calls, zero context degradation. Trained with "environment expansion" across multiple agent frameworks. (AI Revolution — "Elon Musk Just Shocked OpenAI With Grok 5")
- MiniMax M3 teased with sparse attention architecture: 10x faster context processing, 15x faster decoding. Based on GQA (not MLA), operates on real KV cache for better reasoning quality. Expected June. (WorldofAI)
- Gemini 3.5 Pro X-High thinking variant appearing in backend flags. Google also preparing Gemini 3.1 Flash Live with potential voice cloning (model ID: VR-EAP). (WorldofAI)
- MiMO 2.5 Pro gets 99% API price cut — now costs roughly the same as DeepSeek V4 Pro. Xiaomi competing aggressively on efficiency. (WorldofAI)
- Four new Anthropic codenames leaked: tunes, squares, bitboard, claude spaces — pointing toward collaborative workspaces, persistent agent environments, and customizable AI workflows. (WorldofAI)
- DeepSwe — new agentic coding benchmark for realistic software engineering (not scraped from GitHub). GPT 5.5 reportedly already scoring ~70%. (WorldofAI)
Open Source
- DeepSeek researcher publishes 99% AI-written paper. Deli Chen (core contributor to DeepSeek V1-V4, R1, Coder) published a 46-page survey using his autonomous research agent. First draft: 76 minutes. Total: 6 days, 108 rounds, 648K tokens, 2,234 lines of LaTeX, 103 verified references. Proposes a 5-level autonomy taxonomy for research agents (like self-driving levels). Identifies 6 unsolved problems including cognitive loop traps, context limitations, and reproducibility. (AI Revolution)
- Deno open-sources Claw Patrol — an agent firewall for securing AI agent actions. (Better Stack — "Deno Just Open Sourced Their Agent Firewall")
- MiniCPM5-1B reportedly beating Qwen and Gemma 4 at its size class. (Ray Codes — "Destroying Qwen and Gemma 4")
- Hermes Agent trending across multiple channels as a new agentic framework gaining developer traction. (Parker Prompts; Metics Media)
Industry Moves
- SpaceX targeting $1.75T IPO on June 12 — would be the largest IPO in history. $60B Cursor acquisition to complete within 30 days post-IPO. (AI Revolution)
- Figure AI signs commercial deployment deal with Catalyst Brands (J.C. Penney, Aeropostale, Brooks Brothers) — Figure 01 robots deploying in Reno, NV. First real commercial-scale humanoid robotics deployment. (WorldofAI)
- China restricting foreign travel for top AI researchers including at Alibaba and DeepSeek. Government approval now required — previously reserved for defense/nuclear. Triggered by Manus AI's Singapore relocation and brain drain fears. AI talent now treated as national security asset. (Firstpost — "China's AI Panic? Beijing Restricts Travel for Top AI Talent")
- Electrical grid being rebuilt for AI — Bloomberg deep dive (151K views) on infrastructure transformation required for AI data center power demands. (Bloomberg Originals)
Safety & Policy
- Trump AI executive order killed by three phone calls. The most significant AI governance setback of 2026 so far. The order would have created voluntary pre-release vetting — not even mandatory regulation. White House officials "caught off guard." Mythos capabilities specifically cited as the catalyst for wanting regulation. Continuing story from Pope Leo's encyclical (May 25-27 coverage). (MS NOW; Politico reporting)
- Eliezer Yudkowsky long-form interview on catastrophic AI risk. Describes superintelligence designing biological infrastructure "the size of dust mites, stronger than diamond." Notes Geoffrey Hinton (Nobel laureate, inventor of deep learning) puts catastrophe probability at 25-50%. Compares AI industry's denial to leaded gasoline and cigarette companies. 14K views. (Neural Nutshell — "AI Whistleblower WARNS: You Have NO Idea About The AI Wave That Is Coming")
- Brookings Institution panel on regulating AI companion bots as a public health issue — new regulatory framework proposed. (Brookings Institution)
Worth Watching
- "The acceleration is here!" — AI Search (47K views). Two Nature papers on AI autonomous scientific discovery, clearly explained. The CUR-6 leukemia discovery is genuinely remarkable.
- "The 'AI Job Apocalypse' is CANCELLED!" — Wes Roth (19K views). Best analysis of why Silicon Valley is reversing its job displacement narrative, with the Jevons paradox framing.
- "Elon Musk Just Shocked OpenAI With Grok 5" — AI Revolution (47K views). Covers Grok V9/Cursor data, DeepSeek's AI-written paper, and Qwen 3.7 Max's Code Arena breakout.