Anthropic's biggest week ever turned into a security and PR nightmare. Days after launching Fable 5 — its most capable model yet, built with safety and guardrails as headline features — the company was forced to take it offline entirely, for everyone.

The story starts with a creative jailbreak. Researcher Pliny the Liberator (@elder_plinius) used a coordinated "pack hunt" technique: multiple AI agents working in parallel, using Unicode homoglyph substitution (Cyrillic characters replacing Latin ones), long-context narrative smuggling, and decomposition of harmful queries into benign-looking fragments. The model responded with detailed output on stack buffer overflow exploits and sensitive chemical synthesis pathways.

That alone would have been bad enough. But Pliny also published the full Fable 5 system prompt — approximately 120,000 characters of internal safety instructions, classifier logic, fallback behaviors, and runtime configuration that Anthropic had designed to remain confidential. It's one of the largest system prompt leaks in AI industry history.

Why it matters

The government response was unprecedented. The U.S. Commerce Department issued an export control order barring Anthropic from providing Fable 5 and Mythos 5 access to any foreign nationals — including those inside the U.S. Since the company had no practical way to enforce nationality-based restrictions on API and consumer access, it shut both models down for all users.

According to Tom's Hardware, presidential adviser David Sacks stated that Anthropic had been warned in advance about the vulnerability and declined to patch it before the government stepped in.

This raises serious questions for the entire AI industry: Are export controls the right tool for managing frontier model risks? What happens when a private company can't meaningfully control what its own software does in the wild? And how will this affect Anthropic's reported IPO plans? Industry estimates suggest billions of dollars in potential value impact.

Other tech stories making waves

It's not just Anthropic dominating the headlines. SpaceX completed the largest IPO in history last Friday — raising roughly $75 billion at $135 per share. The stock closed up 19% on its first trading day, giving the company a market cap above $2 trillion and making Elon Musk the world's first trillionaire.

Humanoid robotics also took a leap forward. 60 Minutes aired a segment on Boston Dynamics' electric Atlas robot now working at Hyundai's Georgia factory, performing semi-autonomous parts-sorting tasks on the assembly line — a real step from lab demos to industrial reality.

Google released DiffusionGemma 26B-A4B, a new open-source diffusion model, adding to the week's launch wave.

The bottom line

June 2026 is shaping up as a landmark month in AI. The Fable 5 jailbreak holds up a painful mirror to the industry: the most sophisticated safety measures money can buy still didn't stop a coordinated attack, and the government's swift response could reshape the playing field for AI companies worldwide. Meanwhile, SpaceX reminds us that American tech isn't just about AI — the market still wants rockets, space infrastructure, and physical industrial innovation.