The Weekly Waypoint, Issue #22
For years, the question was theoretical: could the US government shut down a frontier AI model? This week, it stopped being theoretical.
The Commerce Department issued an export control directive that forced Anthropic to disable Claude Mythos 5 and Claude Fable 5, its most powerful models, for all non-US users. Within hours, the models went dark. Cybersecurity teams who were using Mythos to find zero-day vulnerabilities lost their tool. European allies scrambled. China's DeepSeek raised $7.4 billion the same week. And the message was clear: the government doesn't need new laws to control AI. It already has the lever.
That wasn't the only chaos. OpenAI's financials leaked showing $34 billion in spending and a $39 billion loss in 2025. SpaceX bought Cursor for $60 billion. Google lost two of its biggest AI researchers in the same week. The AI industry's center of gravity is shifting, and nobody's sure where it lands.
Let's get into it.
The Anthropic Shutdown: First-Ever AI Model Export Ban
The Commerce Department sent Anthropic a letter restricting access to Claude Mythos 5 and Fable 5. Anthropic complied within hours. The models went offline for all foreign nationals. The stated concern: the models could be diverted to foreign military intelligence operations. The trigger, reportedly: a group with ties to China had accessed the models through SK Telecom in Korea.
This is the first time the US government has used export controls to shut down a deployed frontier AI model. Not a future model. Not a research project. A model that was live, being used by cybersecurity teams to find vulnerabilities in critical infrastructure. Mythos had already identified over 23,000 vulnerability candidates, most still unpatched. Now the tool that found them is gone.
What this means: Every AI company now knows the government can pull the plug. No new legislation required. No congressional vote. A letter from Commerce, and your model is dark. If you're building on US-based AI, this is your risk picture.
Cybersecurity Backlash: Security Experts Revolt
Within 48 hours of the shutdown, cybersecurity leaders started pushing back hard. CrowdStrike, Cloudflare, and other security firms reportedly urged the White House to reverse the ban. The argument: Mythos was finding real vulnerabilities that needed fixing. Without it, those holes stay open. The ban helps attackers by blinding defenders.
Senator Ron Wyden reportedly said Mythos "broke into almost every US classified system in hours" during testing, which is either an argument for restricting it or an argument for fixing the systems. Depends on who you ask.
What this means: The security community and the policy community are on a collision course. AI is simultaneously the best defensive tool we have and the thing policymakers fear most. That tension doesn't resolve cleanly.
OpenAI's $34 Billion Burn: IPO Math Gets Hard
The Financial Times leaked OpenAI's 2025 financials: $34 billion in spending, $39 billion in losses. Q1 2026 burn rate: $3.7 billion. Revenue is growing fast, but costs are growing 8x faster year-over-year. The IPO filing is moving forward anyway, with reports suggesting a 2026 timeline.
SoftBank's stock dropped on the news. OpenAI's CFO reportedly flagged the timeline as aggressive. Sam Altman reportedly said he was "0% excited" to run a public company. The IPO market for AI is heating up, but OpenAI's financials make it the riskiest IPO in tech history.
What this means: The biggest AI company in the world is losing more money than any startup in history. The IPO will price future growth against current burn. If revenue doesn't catch up to costs before the IPO window closes, things get interesting.
SpaceX Buys Cursor for $60 Billion
SpaceX agreed to acquire AI coding startup Cursor for $60 billion in stock. The four co-founders, all in their mid-20s, became multibillionaires overnight. This is the largest AI acquisition ever and the clearest signal that Elon Musk is vertically integrating AI into everything SpaceX does.
The logic: Cursor is the best AI coding tool on the market. SpaceX needs world-class software for rockets, satellites, and Starlink. Owning the coding AI means owning the pipeline. Competitors -- Anthropic, OpenAI -- now have a new enemy in the coding space, one with infinite compute ambition and a $400B+ valuation.
What this means: Vertical integration of AI is accelerating. The assumption that the best AI tools come from dedicated AI companies is under pressure. SpaceX, Amazon, Google, Microsoft -- they all want to own the full stack.
DeepSeek Raises $7.4 Billion at $50B Valuation
DeepSeek closed its first external funding round: $7.4 billion at a valuation over $50 billion, making it China's most valuable AI startup. The deal structure was unusual: founder Liang Wenfeng personally contributed $3 billion, maintaining tight control. Investors signed a no-poaching clause -- they can't hire away DeepSeek employees.
This happened the same week the US government restricted Anthropic's models. The timing is not lost on anyone. While America restricts its own AI champions, China's champion just got $7.4 billion richer.
What this means: The AI race is global, and US export controls have a side effect: they strengthen the case for sovereign AI in every country that isn't the US. DeepSeek's funding round is the counter-narrative to American AI dominance.
Talent War: Google Bleeds, OpenAI and Anthropic Poach
Two of Google's biggest AI researchers left this week. Noam Shazeer, co-lead of Gemini and co-inventor of the transformer architecture, is joining OpenAI. John Jumper, Nobel laureate for AlphaFold, is joining Anthropic.
Shazeer is the one that hurts. He helped invent the transformer -- the T in GPT. He was running Gemini. Now he's at the company trying to build GPT's successor. Sam Altman called it "ten years in the making." Google paid $2.7 billion to bring him back from Character.AI two years ago. That investment just walked out the door.
What this means: Google is still a top AI destination, but it's no longer the default. OpenAI and Anthropic are winning the talent war for the most consequential researchers. When the person who invented the transformer leaves for the competition, that's a signal.
Quick Hits
- Visa + OpenAI integration. ChatGPT can now process payments through Visa's network. AI agents that buy things is no longer theoretical.
- Salesforce acquires Fin for $3.6B. AI agent spending spree continues. Salesforce is betting its entire future on Agentforce.
- Anthropic launches $150M Claude Corps. 1,000 fellowships to bring AI to nonprofits. Smart positioning while fighting the government.
- Pew Research: Americans and AI 2026. New survey on public perception of AI. The numbers are worth reading if you build AI products.
- Sanders proposes AI tax. A tax on AI company stocks. Unlikely to pass, but signals where the political wind is blowing.
- China announces AI consumption push. New government measures to integrate AI into consumer products. State-directed AI adoption at scale.
The Takeaway
This week proved three things. The government can shut down your AI model with a letter. The biggest AI company in the world loses more money in a quarter than most companies ever make. And the talent building these systems is now as mobile as it's ever been, flowing toward whoever has the most compelling story (and stock options).
If you're building on AI, the risk equation changed. Your model provider can be shut down. Your competitor can be acquired for $60 billion overnight. Your best engineer can be poached by a company that didn't exist five years ago. Plan accordingly.
Deep Dive This Week
The Kill Switch Era: When Governments Can Turn Off AI, and What It Means for Everyone
The full breakdown of the Anthropic shutdown, the cybersecurity fallout, the OpenAI financials, the SpaceX-Cursor deal, DeepSeek's $7.4B raise, the talent war, and what it all means for your AI strategy. 12 sections. The most consequential week in AI since ChatGPT launched.
Not Pro yet? Upgrade for $6/month
See you next Monday. -James
This week's newsletter was written between bouts of staring at the screen wondering if any of this is real.