Trump says he no longer views Anthropic as a national security threat after G7 meeting with CEO — The Next Web After Commerce Secretary Howard Lutnick imposed export controls on Mythos 5 and Fable 5 — forcing Anthropic to cut off global model access — Trump reversed course following a G7 meeting with CEO Dario Amodei. The restrictions have not been formally rescinded, and both sides are reportedly negotiating shared standards for evaluating AI jailbreaks.

🏛️ Policy & AI Safety

Trump says he no longer views Anthropic as a national security threat after G7 meeting with CEO — The Next Web

The about-face came days after Lutnick sent Dario Amodei a letter restricting Mythos 5 and Fable 5 under export controls, triggered by reports another company had jailbroken Mythos and alarmed the administration. Anthropic cut off access globally to comply; formal rescission of restrictions is still pending as the two sides work toward jailbreak evaluation standards.

The Flat Curve Society — Steve Yegge / Medium

Yegge argues that Anthropic’s public framing of AI as existentially dangerous has a strategic dimension: as capabilities grow exponentially, access to the most capable models will be locked behind government controls, with open models unlikely to advance past Fable class due to compute requirements and policy restrictions. A pointed read on the entanglement of safety narratives and competitive moats.

Why Amazon hates ‘human-in-the-loop’ AI governance — The Register

💰 Funding & Industry Moves

AI chipmaker Groq confirms $650M raise, re-staffs after Nvidia’s $20B not-acqui-hire deal — TechCrunch

Six months after Nvidia licensed Groq’s LPU technology and hired away its CEO and president in a $20B deal, Groq confirmed a $650M raise from existing investors to pivot from chip design to AI inference cloud. New leadership — a COO from xAI/Meta and a fresh CTO/CPO pair — signals a serious second act positioning Groq against Nvidia’s inference dominance.

Google DeepMind bets $75M on AI’s future in Hollywood with A24 deal — TechCrunch

Google DeepMind invested $75M in indie studio A24 in a “first-of-its-kind” research partnership to build AI filmmaking tools alongside working artists — without access to A24’s content library or data. Demis Hassabis: “We believe the best way to develop tools that empower artists is to work directly with them.” A deliberate counterpoint to Hollywood deals framed around cost-cutting.

SpaceX inks compute deal with Reflection AI, an open source AI lab — TechCrunch

Reflection AI will pay $150 million per month beginning July 1 through 2029 for immediate access to NVIDIA GB300 Blackwell chips across SpaceX’s Colossus 2 data center near Memphis. One of the largest compute commitments by an open-source AI lab, it further cements SpaceX as a major neocloud competitor.

Tesla plans to sell modular AI data center hardware called ‘Megapod’ — Electrek — Tesla filed a trademark for a turnkey AI data center building block — servers, networking, power, and cooling in one — that would compete directly with Nvidia’s platform.

Secretive Wall Street Powerhouse Jane Street Seizes the AI Spotlight — WSJ — Hiring 500+ this year, Jane Street is openly courting AI startups and compute deals, a sharp break from its historically low profile.

About Those “Hackquisitions”… — Spyglass — A sharp retrospective on not-acqui-hire deal structures: Noam Shazeer’s departure from Google DeepMind for OpenAI in under two years — just after Google paid $2.7B to bring him back — is exhibit A.

🤖 Models & Talent

ChatGPT market share falls below 50% for the first time — Build Fast With AI

ChatGPT’s share of the global AI assistant market dropped to 46.4%, with Gemini rising to 27.7% and Claude at 10.3%. The sharper metric: Claude leads on paid conversion at 13% — highest in the field — suggesting strong monetization even at lower headline share.

Noam Shazeer departs Google DeepMind for OpenAI as Lead for Architecture Research — Build Fast With AI — The Transformer co-author and Google’s $2.7B re-hire makes his second high-profile departure, this time to lead architecture research at OpenAI.

AWS Weekly Roundup: Grok 4.3 in Bedrock, EC2 price reductions, NY Summit recap — AWS

🛠️ Developer Tools & Infrastructure

Temporary Cloudflare Accounts for AI agents — Cloudflare

wrangler deploy --temporary lets AI coding agents spin up live sites for 60 minutes with zero auth friction, iterate through changes, and optionally claim deployments to make them permanent. This closes a key agentic loop: agents can now verify their own work against a real deployment without human intervention.

Don’t rely on instructions, use Agent Hooks to enforce guardrails — Zarar.dev

Agent Hooks provide deterministic mid-task intervention points rather than post-hoc review, ensuring rules are enforced 100% of the time. The post demonstrates two practical hooks: blocking direct input tags and preventing agents from declaring work done while tests are still failing.

Production-Ready Autonomous Incident Resolution with AWS DevOps Agent (now GA) and Datadog MCP Server — AWS — GA release with automatic Slack/PagerDuty/ServiceNow coordination, cutting incident response from hours to minutes.

Amazon EC2 G7 instances with NVIDIA RTX PRO 4500 Blackwell GPUs now GA — AWS — AWS becomes the first major cloud to offer NVIDIA’s RTX PRO 4500 Blackwell Server Edition GPUs.

GitLab 19.1: AI governance with agent action audit streaming and approval guardrails — GitLab

Jcode — open-source coding agent with “self-dev mode” — GitHub — Claims 10x faster boot than competitors; agents can modify their own source code at runtime.

Cloudflare One Stack: agent-powered Zero Trust deployment automation — Cloudflare

I am dreading our LLM-written incident report future — Surfing Complexity

🔒 Security

Read this before you vibe-code another app — The Verge

The story of “Boomberg” — a vibe-coded tax transparency site that launched with a SQL injection vulnerability its creator didn’t discover for months — anchors a broader look at the widening gap between code that appears to work and code that is actually secure. As AI-assisted development accelerates, the security review step is the one most likely to get skipped.

Lighthouse agentic browsing scoring — Chrome Developers — Google launches deterministic audits to score how well sites are built for AI agent interaction.

When it’s the maintainer who’s AI-pilled — Octet Stream — On using well-supervised agents to accelerate the PR pipeline: turning open issues into commits within hours.

Generated by claude-sonnet-4-6 · 2026-06-22T10:00:00Z