Happy Friday, fam 👋

This was the week the AI map got redrawn three times in 72 hours: Musk turned IPO cash into a $60B coding empire, Washington reached into the cloud and switched off Anthropic’s two best models, and — the same week — Anthropic walked into Seoul and signed Samsung and LG. And Woody and his friends fighting off AI. Buckle up.

Headless browsers get flagged by bot-detection the moment they load. Human Browser (by Virix Labs) is the opposite: a real, cloud-based Chrome session on a residential IP that behaves like a person — Bezier-curve mouse movement, human-speed typing, and natural scrolling with reading pauses.

Any agent that speaks HTTP — Claude, Cursor, Codex, Gemini — can hand it a task over the A2A protocol or MCP, watch the session live, and get the result back. It’s built for agent developers who need to work the open web without tripping every alarm.

Four days after the biggest IPO in Wall Street history, SpaceX is acquiring Cursor-maker Anysphere for $60 billion in an all-stock deal, with closing slated for Q3 pending regulatory clearance. The structure: roughly 3.4% dilution to SpaceX shareholders, no cash component, and a planned merger that pulls Cursor under the same xAI/SpaceX umbrella as Grok and the Colossus supercluster. Musk now has the GPUs, the chatbot, the rocket pipeline, and the editor where the world's developers actually write code.

The price tag tells you why Wall Street didn't blink. Cursor crossed $1B in annualized revenue in November and is reportedly running near $4B ARR today — roughly 4× in seven months and the fastest climb from $1M to $500M ARR any SaaS company has ever recorded. One survey put Cursor inside 64% of the Fortune 500. The strategic prize isn't the revenue; it's the surface area. Whoever owns the editor owns the model-routing decision millions of developers make every day, and Cursor's neutrality-by-default — Claude, GPT, and its own Composer all available in one box — is exactly the chokepoint Anthropic and OpenAI cannot let drift to a third party. They just lost it anyway.

On June 12, the Trump administration gave Anthropic a binary: patch a jailbreak that the Commerce Department's national-security advisors had flagged, or accept a full export-control directive. Anthropic, which characterized the issue as a "narrow, non-universal jailbreak" rather than a systemic flaw, declined to recall the models — and the Commerce Department issued the order anyway. Within hours, Anthropic abruptly cut all foreign access to Fable 5 and Mythos 5, its two flagship frontier models. As of Wednesday June 17, in-person talks between Anthropic engineers and Commerce officials are still ongoing, with no deal struck and no restoration date set.

The directive is unprecedented: it is the first time the US government has unilaterally pulled an American AI lab's frontier models off the global market. And it didn't happen in a vacuum. Anthropic and the Trump administration have been on collision course since February, when the company walked away from a Pentagon contract rather than allow Claude to be used for autonomous lethal targeting and mass civilian surveillance. The Fable 5 jailbreak — exploitable but, by Anthropic's own technical assessment, not catastrophic — was the lever, not the cause. The pattern this sets is what's worth watching: when an American lab and the American government disagree on red lines, the lab loses access to roughly half its addressable market overnight.

In the exact 72 hours the Commerce Department was shutting down Fable 5 abroad, Anthropic was opening its Seoul office — its third in Asia-Pacific — and announcing the largest single-day Korean enterprise wave in its history. NAVER committed to deploying Claude Code across its full engineering organization. Samsung SDS rolled Claude Cowork and Claude Code out to Samsung Electronics staff. LG CNS put Claude in front of thousands of LG employees with an expansion path across the broader LG Group. Nexon wired Claude Code into its live-service game pipeline. Hanwha Solutions went live via AWS Bedrock. Five marquee logos in one announcement. The Seoul office is led by KiYoung Choi, formerly General Manager of Snowflake Korea — a signal that the Korea bet is a serious enterprise play, not a representative-office flag-plant.

The contrast writes itself. While Washington was telling Anthropic which markets it couldn't sell into, Anthropic's enterprise team was busy locking up the Korean market that very same week. Claude Sonnet and Haiku remain available globally — only the very top of the model stack is gated — and Korean customers signing 18-month enterprise deals are buying capabilities, support, and a roadmap, not just a single model. The deeper lesson: the policy fight and the commercial fight are running on two different clocks, and the commercial clock moves faster.

GitHub vs. itself. Microsoft’s AI coding agents — many of them powered by Microsoft’s own Copilot — overwhelmed Microsoft’s own GitHub to the point that, on June 16, Microsoft confirmed it’s routing GitHub traffic through Amazon Web Services. The richest company in the world is running its developer platform on its biggest competitor’s cloud because its AI couldn’t keep up with its AI. Beautiful.

Runner-up: the jailbreaker known as Pliny the Liberator posted a coordinated multi-agent Unicode/homoglyph “pack hunt” on Fable 5 on June 10. Three days later the US government export-controlled the model. He’s now “the guy who got Anthropic export-controlled” — at least on X.

Consolidation just hit AI coding. Here’s the field worth knowing right now:

This week’s main character is Michael Truell, co-founder and CEO of Anysphere — the company behind Cursor. He and his MIT co-founders turned a fork of VS Code into the fastest-growing developer tool in recent memory, and just agreed to fold it into SpaceX for $60 billion. Whatever you make of the deal, Truell built the editor your team probably argued about this week — and did it barely out of school. Cursor is the tool the industry is now reorganizing around.

Our AI Graveyard keeps growing. Four fresh headstones this week:

The Fable 5 export control is the first time the US government has unilaterally pulled an American lab’s frontier model off the foreign market. Not a fine, not a hearing — an off switch.

Pair it with February, when Anthropic refused to support lethal autonomous and surveillance use for the Pentagon. A lab drew a red line on how its models could be used; months later the government drew its own red line on where they could go.

The structural lesson is uncomfortable: when a country and its leading AI lab disagree on red lines, the country wins. Compute, export law, and national-security authority give the state leverage no cap table can match.

And every restriction aimed at a rival also hands that rival a script. Open-weight players were pitching “this is why closed models are fragile” within hours. Security by subtraction has a cost — and it isn’t always paid by the intended target.

Worth a spot on the calendar:

June (last week of the month)

July

August

Building something interesting, or know a tool that belongs in the directory (or the Graveyard)? Hit reply, or email [email protected] — we read everything.

Enjoy Toy Story 5 this weekend with the gang fighting off AI!!!

— The ToolDirectory.AI team