Treasury calls about Claude, CoreWeave's $21B Meta bet, and OpenAI's leaked war memo
Tue, Apr 14, 2026 · 10 stories
Anthropic had the week. Their new Mythos model is so good at finding cybersecurity vulnerabilities that the Treasury Secretary and Fed Chair convened an urgent meeting with Wall Street CEOs to warn them. They locked in a multibillion-dollar CoreWeave compute contract the same week Meta signed its own $21B deal. And a leaked OpenAI internal memo made it clear who they're scared of: the whole strategy now is building moats to stop customers switching to Claude.
Top Stories
Bloomberg Technology
Treasury Secretary Scott Bessent and Fed Chair Jerome Powell urgently convened Wall Street CEOs to warn of cybersecurity risks from Anthropic's new Mythos model, which has advanced vulnerability-detection capabilities. Anthropic restricted the release to vetted parties only; ECB President Lagarde publicly praised the cautious approach while the IMF urged governments to stay on the AI risk frontier.
Why this matters:
This is the week frontier AI capability officially collided with systemic financial policy. When central bankers summon bank CEOs over a single model release, the era of "move fast and let product teams figure out safety" is over for any AI company selling into regulated industries. Expect every major financial regulator to now demand pre-release access, and watch which labs position that as a moat vs. a tax.
Bloomberg Technology
CoreWeave signed a $21 billion compute supply agreement with Meta through 2032 and a separate multibillion-dollar contract with Anthropic, then immediately tapped the junk bond market to finance capacity expansion. Meta simultaneously confirmed it rented Norwegian Stargate data center capacity that OpenAI had originally reserved.
Why this matters:
Compute is being allocated on multi-year forward contracts now, not spot purchases, and the financing stack for those contracts is moving into high-yield debt. For VCs, this is the first clear signal that GPU supply is a seller's market through at least 2028. For founders, it means your eventual hyperscaler pricing is being set by these seven-year contracts, not by the "cheap inference" narrative.
The Verge
OpenAI CRO Denise Dresser's internal memo, leaked to The Verge, outlined explicit plans to "lock in" users and build enterprise moats against Anthropic. The same week, OpenAI launched a $100/mo ChatGPT Pro tier with 5x Codex usage (matching Claude Max pricing), acquired personal-finance startup Hiro, and paused its UK Stargate effort citing energy costs.
Why this matters:
An internal "build moats, prevent switching" memo is what a company writes when it's losing switchers. The $100 tier is a direct Claude Max response, and the simultaneous Hiro acquisition plus Cloudflare Agent Cloud integration tells you OpenAI is racing to fill verticals before Anthropic's enterprise momentum compounds. For founders picking a default model, this is the clearest sign yet that vendor lock-in will get worse before it gets better.
Quick Hits
TechCrunch AI
Signals ChatGPT's push into financial advisory; expect consumer-fintech ripple effects.
The Verge
Zuckerberg's rebuilt AI group shipped its first model; Meta AI app climbed App Store from No. 57 to No. 5 in 48 hours.
OpenAI Blog
Enterprises can now deploy OpenAI agents on Cloudflare's edge with a single SDK, compressing weeks of agent-deployment work into hours.
DeepMind Blog
Google's updated robotics model improves multi-view spatial reasoning for real-world manipulation tasks.
MIT Technology Review
Annual data-driven snapshot of the AI industry; essential quarterly bookmark for anyone citing stats.
Hacker News
Torvalds merged formal guidance on using Claude, Copilot, and other AI tools for kernel patches, with disclosure requirements.
LangChain Blog
Model-agnostic production agent platform, positioned directly against Anthropic's managed offering.
