SpaceX Absorbs Cursor as AI Market Battles “Tulipmania” and OpenAI Stumbles — 2026-06-16#

Highlights#

The AI community is reeling from SpaceX’s acquisition of Cursor, an applied AI milestone that has pushed SpaceX’s valuation to a bubble-like $2.5 trillion and drawn comparisons to historical financial manias. Meanwhile, regulatory panic has set in as the White House attempts to rein in Anthropic’s frontier models, even as former titan OpenAI faces dwindling market share and a wandering eye from key partner Microsoft.

Top Stories#

• SpaceX Acquires Cursor, Valuation Moons: SpaceX’s merger with applied AI standout Cursor has sent its valuation soaring past Apple and Amazon to $2.5 trillion, sparking widespread “Tulipmania” warnings across the sector. Critics argue this 1929-style speculation vastly disconnects from SpaceX’s $18 billion revenue, though some predict SpaceX AI will quickly rival Google and OpenAI.
• White House Torpedoes Anthropic’s Claude Models: The Trump administration has halted Anthropic’s Fable 5 and Mythos 5 over “jailbreak” concerns that security experts dismiss as trivial, denying even UK Prime Minister Keir Starmer an exemption. Rumors suggest the administration may use the restrictions as leverage to extract financial stakes in major AI labs.
• OpenAI’s Market Share Dips Below 50%: OpenAI is bleeding market dominance as Google scales, pushing ChatGPT’s market share below the 50% threshold. Microsoft is now publicly exploring cheaper alternatives like DeepSeek, compounding industry fears over OpenAI’s projected $38 billion loss for 2025.
• GLM-5.2 Brings Frontier Intelligence to Open Weights: GLM-5.2 has launched with a 1 million token context window and MIT-licensed open weights. Running effectively on Mac Studios, it pushes the limits of agentic and coding tasks to rival closed-source frontier models.

Articles Worth Reading#

The Fall of Fable 5: Export Controls on Trivialities Security researcher Katie Moussouris broke down the guardrail bypass paper that supposedly got Anthropic’s Fable 5 restricted. Industry experts are pointing out that the “dangerous” jailbreak triggering the White House shutdown essentially boiled down to a benign prompt asking the model to “fix this code”. The incident is drawing sharp criticism over the application of heavy-handed export controls on software rather than focusing on genuine cyber resilience.

Temporal Difference in Vision (TDV): Causality Over Assumptions A new representation learning paradigm called TDV challenges the traditional necessity of augmentations, masking, and cropping for computer vision. By relying purely on causality as its foundational assumption, TDV matches state-of-the-art models like DINO and iBOT on dense spatial tasks. This research signals a pivotal shift in AI scaling, proving that relying on weaker assumptions often yields better representation learning results as data scales up.

Hermes Agent Gains Autonomous Commerce Skills via Stripe Nous Research has partnered with Stripe to embed a full suite of financial capabilities into the Hermes Agent. Agents can now autonomously execute purchases, pay for per-call APIs, and provision SaaS environments. The integration features configurable safety limits on every action, highlighting the rapid acceleration toward truly autonomous, financially active agentic workflows.