2026-05-22 :: AI DAILY DIGEST #

Trump shelves the AI executive order, Anthropic locks up $15B/year of Musk's compute, and EU–Anthropic talks stall.

📊 TODAY: 18 stories · 16 sources · 🟡 +0.1 sentiment · 🔥 6 cross-source · TOP MENTION: Anthropic ×7
🏷️ THEMES: policy×6, funding×5, agents×4, enterprise×4, multimodal×3
📈 MARKET PULSE: "Trump orders federal review of AI model releases by May 31" 8.5% (▼73.5pp) · 5 AI markets tracked
📉 7D SENTIMENT: ▆▅▅▅▅▅▅ (oldest → today)

⚡ TL;DR #

1. 7 🔥 🛡️ 🟡 Trump postpones the AI security executive order, citing competition with China. White House pulled a draft order that would have tightened model-release review after internal pushback that it would "get in the way" of US labs. (Reuters, FT, TechCrunch) ¶
2. 7 🔥 🟢 Anthropic to pay ~$15B/year for SpaceX/Colossus compute capacity. Multi-year deal positions Anthropic ahead of an IPO and ties it to Musk's data-center buildout — strange bedfellows given the Altman/Musk fight. (The Verge, Reuters) ¶
3. 6 🛡️ 🔴 EU–Anthropic talks over Mythos concerns stall, Spain says. European regulators and Anthropic are deadlocked over the Mythos model's deployment terms; Spain confirmed no movement in the latest round. (Bloomberg) ¶
4. 6 🔥 🟢 Spotify + Universal Music strike fan-made AI cover/remix deal. First major-label framework allowing licensed AI covers, plus an "AI Studio" daily-podcast agent. Sets a template the rest of the labels will have to answer. (TechCrunch, The Verge, FT) ¶

🧠 Models & Releases #

2 stories · 2 sources · 🟢 +0.3 sentiment

• 5 🟢 🏷️ multimodal, enterprise Anthropic's Code with Claude event sketches the next coding workflow. MIT Tech Review's read: agentic IDE behavior, longer-horizon tasks, and a clear signal that "vibe coding" is becoming an industry posture, not a meme. Sources: MIT Tech Review
• 4 🌱 🟢 🏷️ opensource, code Simon Willison ships llm-gemini 0.32 and Datasette Agent. Plugin updates plus a new agent harness that lets Datasette query itself via tool calls. Sources: Simon Willison, Simon Willison

🔬 Research #

2 stories · 2 sources · 🟡 +0.0 sentiment

• 4 🟡 🏷️ science, multimodal Google I/O reframes "AI-driven science" — less moonshot, more workflow. MIT TR argues the path to AI in science is narrower than promised: better lit-review and hypothesis tooling, not autonomous discovery. Sources: MIT Tech Review
• 4 🟡 🏷️ alignment, safety Zvi: AI #169 — "New Knowledge". Roundup covers model behavior shifts, safety-bill politics, and a long take on the Trump EO postponement. Sources: Zvi Mowshowitz

🛡️ Responsible AI, Safety & Policy #

3 stories · 5 sources · 🔴 -0.2 sentiment

• 7 🔥 🛡️ 🟡 ▤×3 🏷️ policy, safety Trump postpones AI executive order. Industry got the delay it asked for; safety advocates lost a draft that would have required pre-release federal review. China-competition framing replaces the prior "secure by default" language. Sources: Reuters, FT, TechCrunch
• 6 🛡️ 🔴 🏷️ policy, safety EU–Anthropic Mythos talks stalled, Spain says. Regulatory friction over a frontier model's deployment terms; Spain publicly confirmed the impasse. Sources: Bloomberg
• 4 🛡️ 🔴 🏷️ policy, enterprise StanChart CEO walks back AI remarks after regulator queries. Bank's CEO apologized for "upset caused" by comments framing AI as a near-term headcount tool; UK regulators followed up. Sources: Reuters, Bloomberg

🎨 Cool Projects & Novel Applications #

3 stories · 3 sources · 🟢 +0.3 sentiment

• 4 🎨 🟢 🏷️ multimodal, art Polyend's "Endless" AI guitar pedal lets you roll your own effects. A standalone hardware pedal that synthesizes new effects on-device from text prompts. Niche, but the most genuinely indie hardware-AI thing this week. Sources: The Verge
• 4 🎨 🟡 🏷️ multimodal, video AI video moves beyond "clip slop". Verge column on Luma/Hollywood pipelines — short-form AI video is starting to look like staged production, not random montage. Sources: The Verge
• 3 🎨 🟢 🏷️ multimodal I cloned myself with Gemini's AI avatar tool. Wired test of Gemini's avatar feature — unnervingly accurate likeness, weird in the way you'd expect. Sources: Wired

💰 Industry & Funding #

5 stories · 6 sources · 🟢 +0.4 sentiment

• 7 🔥 🟢 ▤×2 🏷️ funding, hardware Anthropic locks in ~$15B/year SpaceX Colossus compute. Multi-year capacity guarantee ahead of an IPO; complements separate reporting on Anthropic talks to use Microsoft's custom AI chips. Sources: The Verge, Reuters
• 5 🟢 🏷️ funding, code Modal Labs valued at $4.65B as AI-coding infra heats up. Exclusive Reuters: serverless GPU runtime now a unicorn-plus, riding the agentic-coding wave. Sources: Reuters
• 4 🟢 🏷️ funding, agents Hark raises $700M Series A for a "universal AI interface". Secretive Series A; the price implies an agentic-UI thesis with marquee backers. Sources: TechCrunch
• 5 🟢 🏷️ enterprise, agents JPMorgan rolls out AI tools across investment banking globally. Senior banker confirms firmwide deployment in pitch-prep and modeling workflows. Sources: Reuters
• 5 🟢 🏷️ hardware, robotics Kawasaki Heavy + Nvidia plan Silicon Valley robotics center. Joint robotics lab tied to Nvidia's industrial-AI stack; KHI shares rallied on the news. Sources: Reuters, Bloomberg

🛠️ Tools & Demos #

2 stories · 2 sources · 🟢 +0.3 sentiment

• 5 🟢 🏷️ enterprise, apps OpenAI x AdventHealth — whole-person care deployment. Major US health system standardizes on OpenAI for clinical-summary and patient-comms workflows. Sources: OpenAI
• 4 🟢 🏷️ multimodal, voice Spotify ships AI Q&A and briefing generation for podcasts. Per-episode Q&A and auto-briefings, plus an AI Studio agent that builds a personalized daily podcast. Sources: TechCrunch, The Verge

🌱 Open Source & Emerging #

2 stories · 2 sources · 🟢 +0.4 sentiment

• 5 🌱 🟢 🏷️ opensource, science DeepMind launches Asia-Pacific Accelerator for environmental-risk projects. Free compute + DeepMind mentorship for APAC teams working on climate, biodiversity, and disaster-resilience models. Sources: DeepMind
• 4 🌱 🟢 🏷️ opensource, agents Datasette Agent + llm-gemini 0.32 [via Kagi]. Simon Willison's Datasette gets a self-querying agent; llm-gemini plugin updated for Gemini 2.x. Tight, working open-source release. Sources: Simon Willison

📈 Prediction Markets #

5 AI markets tracked · biggest mover dominates

• "Trump orders federal review of AI model releases by May 31?" Yes 8.5% (▼73.5pp) · $135K vol · Polymarket — collapsed on today's EO postponement.
• "Will OpenAI's market cap be $1.25T–$1.5T at IPO close?" Yes 11.6% (▼7.1pp) · $502K vol · Polymarket
• "Will Google Gemini score ≥55% on Humanity's Last Exam?" Yes 33.5% (▲6.5pp) · $63K vol · Polymarket
• "OpenAI $1T+ IPO before 2027?" Yes 67.5% (▼4.5pp) · $273K vol · Polymarket
• "US enacts AI safety bill before 2027?" Yes 49.5% (▲4.0pp) · $99K vol · Polymarket

💬 Discourse #

Bluesky

• @chrisdeleon.bsky.social on Bluesky (472♥, 136🔁) — "congratulations to every indie game developer I know who has a more profitable business than OpenAI" — the joke that captures the week's mood about AI economics.
• @astrra.space on Bluesky (62♥, 4🔁) — riff on framing an LLM as "a translation model that translates from question to answer" — useful intuition pump for the architecture debate.