Friday delivered two seismic events that will define the summer. At 5 PM, the US Commerce Department imposed export controls on Anthropic's most advanced AI models — forcing a global shutdown of Claude Fable 5 and Mythos 5 and igniting a firestorm about centralized AI control, with open-source advocates calling it a massive catalyst for decentralized alternatives. Hours earlier, SpaceX's historic IPO ($135 → $160.95, $2T+ market cap) made Elon Musk the world's first trillionaire and 4,000+ employees overnight millionaires. Coinbase simultaneously launched "Coinbase for Agents," connecting AI models directly to user accounts for autonomous trading and payments — while Ripple's Brad Garlinghouse publicly declined to connect any agent to his own accounts, underscoring the trust gap. A new academic study showing LLM hacking agents hitting 69.3% pentest success rates adds urgency to Microsoft Build's announcement that OS-enforced agent boundaries (already used by OpenClaw on Windows) are becoming standard. The weekend's twin stories — AI being locked down by governments and AI being unleashed to trade money — are the same story: the agent economy's regulatory and financial infrastructure is being built in real time.
The Commerce Department imposed export controls on Anthropic's most advanced models Friday evening, forcing a global shutdown. Anthropic disputes the rationale — a claimed jailbreak that the company calls already-known minor bugs — but all users worldwide lost access. Fallback: Opus 4.8 and Sonnet 4.6. No restoration timeline.
SpaceX shares jumped 19% on Nasdaq debut Friday, closing at $160.95. The $75 billion raise shattered Saudi Aramco's record. Intraday high of $168.75 pushed market cap to ~$2.21 trillion — briefly the 6th most valuable US company. 4,000+ employees became millionaires.
The new product connects AI agents (ChatGPT, Claude) directly to user accounts for autonomous trading, real-time market data, and payments. Builds on 2024's AgentKit; the x402 protocol already processed 50M+ machine-to-machine transactions. Gasless trading on Base. Brad Garlinghouse publicly declined to connect his own accounts — "protections and structure need to come first."
A new academic evaluation tested LLM-powered agents across 300 controlled servers. The best achieved 69.3% success — confirming that offensive security automation is advancing rapidly. Runs parallel to new research showing OpenClaw agents susceptible to phishing and supply-chain attacks.
Microsoft announced MXC (Mutual Execution Contracts) — define security requirements once and Windows enforces them everywhere agents run. Already deployed with OpenClaw on Windows for multi-step workflows inside OS-enforced boundaries. NVIDIA's OpenShell secure runtime adds policy management, inference routing, and PII obfuscation.
Selected posts from the edition, with links to the originals.
"Can't put the genie back in the GPU."
View post on X ↗"If, when you say regulation, you mean the dead and clammy hand of the commissar—the gentleman who has never in his life built a single thing, drafting rules to govern a thing he cannot define, to be enforced by men who cannot read them; if you mean the form in triplicate, the..."
View post on X ↗"How can Anthropic's IPO proceed in this environment?"
View post on X ↗"According to Community Notes: Amazon researchers did the jailbreak research on Mythos but WSJ never says Amazon reported it to Commerce Dept — that's Theo's inference. Also Anthropic disputes the 'jailbreak,' calling it already-known minor bugs. And Project Glasswing's whole..."
View post on X ↗"ok the government basically just turned off Claude's best AI model overnight. for everyone. over an email they sent at 5pm on a friday lol. and what kills me is there's no good guy here. like yeah it's scary they can flip a kill switch on a product millions use with no..."
View post on X ↗"This is a massive catalyst for open-source AI. Why build your ecosystem on a proprietary model that can be revoked overnight? Top open-source models nearly match proprietary performance, but guarantee something far better: FULL CONTROL."
View post on X ↗"LATEST: 🤖 Brad Garlinghouse says he's 'not going to connect an AI agent' to his brokerage or checking account just yet, arguing that protections and structure need to come first."
View post on X ↗"The AI price war has begun. Startups and tech giants alike are mixing and matching AI models to avoid the premium prices charged by industry leaders."
View post on X ↗"banthropic launching opus 4.9 next week"
View post on X ↗"6 Dangers of Autonomous AI Agents⚠️ Running self-hosted agent frameworks like OpenClaw directly on your local workstation is an operational security nightmare. 🖥️🚨 IBM Security just released a brutal..."
View post on X ↗