Retter Reboot #25

The Retter Reboot : Your weekly dose of ai & tech news

1. SpaceX Buys Cursor for $60B — the Biggest Startup Acquisition Ever

On June 16, just days after its record-breaking IPO, SpaceX agreed to acquire AI coding startup Cursor (Anysphere) in a $60 billion all-stock deal — the largest acquisition of a venture-backed startup on record. The goal is to help SpaceX's AI division, built around the xAI unit it merged with in February, catch up to Anthropic and OpenAI on coding tools; the two have reportedly been jointly training a model on xAI's Colossus supercomputer, expected to ship in both Cursor and Grok Build. SpaceX stock jumped roughly 8% on the news, pushing its market cap past $2.7 trillion, ahead of Amazon and Meta. Cursor, an OpenAI-accelerator alum founded in 2022, was doing about $2.6B in annualized revenue and had been set to raise at a $50B valuation before SpaceX activated a buyout option struck in April. The bigger picture: nearly every major AI coding tool is now owned by an incumbent — Copilot (Microsoft), Codex (OpenAI), Claude Code (Anthropic), and now Cursor and Grok Build (SpaceX). The independent era of AI coding tools is effectively over.

Source: TechCrunch

2. Claude Fable 5 Jailbroken in 48 Hours — System Prompt Leaked

Anthropic launched Fable 5 on June 9 as its most resilient model ever, citing 1,000+ hours of bug-bounty testing that found no universal jailbreak. Within two days, red-teamer Pliny the Liberator bypassed it. His "pack hunt" used coordinated multi-agent tactics — breaking dangerous requests into innocent fragments, Unicode character substitution, and long-context smuggling — to get x86 Linux stack-overflow exploit guidance and a meth-synthesis pathway, then leaked the model's roughly 120,000-character system prompt to GitHub. The real lesson is architectural: Fable 5 shares a base with the restricted Mythos 5 and relies on a classifier that silently reroutes risky prompts to a weaker model. A keyword-and-category filter bolted in front of a powerful model is a blunt instrument. Anthropic disputes the break but separately apologized for over-aggressively degrading the model for legitimate researchers.

Source: Cyber Security News

3. Nadella Warns AI Could "Hollow Out" Entire Industries

In a June 14 essay, Microsoft CEO Satya Nadella argued the simple AI playbook — build the biggest model, win — is about to break. His warning: a world where every company cedes value to a few models that absorb everything they see is one "the political economy will simply not tolerate." His proposed fix is a frontier ecosystem rather than a single frontier model, where every organization owns a "learning loop" that captures its own institutional knowledge through private data and evals, so swapping out a general-purpose model stays easy. It's a notable message from OpenAI's largest backer, landing against Microsoft's ~$190B 2026 capex and a stock down roughly 20% this year. For anyone building on AI, the takeaway is sharp: own your data and your evaluation loop, or rent your competitiveness from someone else.

Source: VentureBeat

4. Japan's Three Megabanks Confirmed for Claude Mythos Access

On June 16, Japan's Finance Minister Satsuki Katayama confirmed that the government and the country's three megabanks — MUFG, SMBC, and Mizuho — will gain access to Claude Mythos, Anthropic's restricted vulnerability-hunting model available through Project Glasswing. It's the first confirmed Glasswing deployment to non-US financial institutions; JPMorgan had been the original banking partner. Mythos sits above the Opus tier and is built for autonomous detection of deep software vulnerabilities rather than general chat. Japan's FSA and the Bank of Japan are convening a public-private working group to manage the dual-use risk: the same model that hardens a bank's defenses could, in the wrong hands, automate attacks against it.

Source: Unrot.co

5. Google Ships "Managed Agents" in the Gemini API

Google quietly shipped one of its more consequential developer features this month: Managed Agents in the Gemini API, now in public preview. Developers can build and deploy autonomous, stateful agents that run inside secure, isolated, Google-hosted Linux sandbox environments — no need to provision or babysit your own execution infrastructure. It's Google's direct answer to the agent-execution layer that Anthropic (Claude Code) and OpenAI (Codex) have used to pull ahead in agentic workflows, and it lowers the barrier for teams that want production agents without standing up their own sandboxed runtimes.

Source: Google AI for Developers

6. US Government Bans Fable 5 and Mythos 5 — First Export Control on an AI Model

In an unprecedented move, the US Commerce Department issued an export-control directive on June 12 ordering Anthropic to suspend all access to Claude Fable 5 and Mythos 5 by any foreign national anywhere in the world — including Anthropic's own foreign-national employees. Because it cannot verify nationality in real time, Anthropic abruptly disabled both models for every customer globally; its other models are unaffected. It's the first time the US has applied export controls to an AI model itself rather than to the chips and hardware behind it, treating a model like a controlled strategic resource. The reported trigger was the model's vulnerability-finding capability and the recent jailbreak. Anthropic publicly disagrees, arguing the jailbreak is narrow and already replicable with other public models, and warning that this standard would effectively halt all frontier deployments. The fallout was immediate: AWS, Google Cloud, Microsoft Foundry, and GitHub Copilot pulled Fable 5, and Glasswing partners in South Korea, the UK, and other allies lost Mythos 5 — handing every sovereign-AI advocate a powerful new talking point.

Source: TIME

7. Midjourney's Surprise Pivot: a Full-Body Medical Scanner in a Spa

In the week's biggest left-field move, Midjourney — the AI image lab — announced a new division, Midjourney Medical, and its first hardware product on June 18: the Midjourney Scanner, a full-body ultrasonic CT device that rings the body with hundreds of thousands of transducers (built on licensed Butterfly Network chips), uses no radiation, and aims to produce a 3D scan in about 60 seconds. Founder David Holz called it the first new whole-body imaging method in 50 years and plans to install ten of them in a Midjourney Spa in San Francisco's Union Square by late 2027, targeting a billion scans a month by 2031. The caveats are large: the prototype currently takes around 20 minutes, only about a dozen people have been scanned, and it has no FDA diagnostic clearance — so it launches as a wellness service, not a medical tool. It fits a broader pattern of AI labs racing into hardware (see OpenAI's robotics and wearable efforts): software can be copied; a physical moat cannot.

Source: The Verge / gagadget

Jun 19, 2026