A stunning export control order halts Anthropic's latest models, SpaceX makes a $60B AI bet, and DeepSeek secures a massive $7.4B war chest.

Last week's edition asked how much of your AI stack you actually control. This week delivered the answer. On June 12, the U.S. Commerce Department ordered Anthropic to disable Claude Fable 5 and Mythos 5 globally — just 72 hours after launch — citing a national security jailbreak that Anthropic publicly disputed. Every customer on the planet lost access simultaneously, not because the models underperformed, but because a government directive made them radioactive overnight.

At the same moment, SpaceX was closing a $60 billion deal to buy Cursor, DeepSeek was locking up $7.4 billion with a deal structure designed to keep Western investors at arm's length, and Salesforce was spending $3.6 billion to acquire the AI customer service layer it couldn't build fast enough. The capital is still flowing. The geopolitics are tightening around it. The real story this week isn't which model is best — it's that the gap between capability and dependability just became the most important thing on your AI risk register.

On June 12, just 72 hours after launching its highly anticipated Mythos-class models, Anthropic was ordered by the U.S. Commerce Department to suspend all access to Claude Fable 5 and Mythos 5 by any foreign national. To comply with the export control directive, Anthropic abruptly disabled the models for all customers worldwide. The government cited national security concerns, reportedly stemming from a potential "jailbreak" that could bypass the models' cybersecurity safeguards.

Anthropic strongly disputed the severity of the jailbreak, stating the capabilities demonstrated were already available in other public models like OpenAI's GPT-5.5. CEO Dario Amodei and the administration clashed over the response, with the White House claiming Anthropic refused to fix the issue, while the company criticized the government's opaque process. The move represents an unprecedented use of export controls against a commercially deployed AI product, setting a stark precedent for how quickly the government can intervene in the AI market.

The shutdown lands at a highly sensitive moment for Anthropic, which recently filed a confidential IPO prospectus revealing a staggering $47 billion revenue run rate and a $965 billion valuation. For enterprise leaders, this incident introduces a new vector of risk: the possibility that critical AI infrastructure could be pulled offline overnight by government mandate, regardless of commercial agreements.

Anthropic's reported valuation in its confidential IPO prospectus, driven by an explosive $47 billion annualized revenue run rate.

Source: CNBC

SpaceX is making a massive play for developer productivity, agreeing to acquire AI coding startup Cursor for $60 billion in an all-stock transaction. The deal, expected to close in Q3 2026, comes just days after SpaceX's blockbuster IPO and represents a 3.4% dilution of its valuation. For engineering teams, this signals that AI-assisted coding is becoming critical infrastructure, with Cursor's market share dropping to 26% recently as it faces fierce competition from Anthropic and OpenAI.

Chinese AI startup DeepSeek has secured over $7.4 billion in its first external funding round, pushing its valuation beyond $50 billion. The deal features an unusual structure designed to preserve founder Liang Wenfeng's control, requiring investors to put capital into a limited partnership rather than the company itself, complete with a five-year lock-up and no voting rights. This massive war chest solidifies DeepSeek's position as China's national AI champion and a formidable global competitor.

Salesforce is accelerating its push into autonomous tech by acquiring AI customer service platform Fin (formerly Intercom) for approximately $3.6 billion. Fin's AI agent, powered by its proprietary Apex model, can resolve queries across chat, email, WhatsApp, and Slack. The acquisition will bolster Salesforce's Agentforce platform, providing enterprises with more robust tools to automate customer interactions at scale.

The CEOs of OpenAI (Sam Altman), Anthropic (Dario Amodei), and Google DeepMind (Demis Hassabis) joined world leaders at the G7 summit in France to discuss AI infrastructure, sovereignty, and frontier risks. The high-level meeting underscores the growing geopolitical influence of tech executives, as governments realize they need the cooperation of private labs to establish credible AI commitments and global baselines.

Startup Probably raised $9 million in seed funding from Andreessen Horowitz to build a more reliable kind of AI. Their approach uses a "data science mech suit" that checks an LLM's first-pass answers against a deterministic validator system, aiming for 99.99% accuracy. By refining the context and reducing ambiguity, Probably allows complex tasks to run on smaller, cheaper models, offering a compelling solution for precision-sensitive enterprise use cases.

Databricks Genie One — An agentic AI coworker for business teams that analyzes company data, automates workflows, and turns business context into actions, providing a governed AI assistant within the Databricks environment.

Higgsfield Supercomputer 2.0 — Built on NVIDIA's Agent Toolkit, this enterprise-grade marketing product provides brands with an autonomous system to create, launch, and optimize ads and video campaigns from a single chat interface.

Probably — A data science tool designed to prevent hallucinations by running LLM outputs through a deterministic validator. It provides fast, accurate answers with citations and audit trails, ideal for precision-sensitive tasks like accounting or medical services.

This analysis unpacks the broader implications of the U.S. government's decision to force Anthropic's models offline. It explores the precedent set for government intervention in commercially deployed AI, the debate over "jailbreaks" and security safeguards, and what enterprise security teams need to know when relying on frontier models that could be subjected to sudden export controls.

If you only do one thing this week: audit your organization's reliance on single AI providers. The Anthropic shutdown proves that even the most capable models can vanish overnight due to regulatory action, making a multi-model strategy essential for business continuity.