SpaceX's Cursor Acquisition Announcement Erases $600B as AI Coding M&A Rattles the Market

According to Forbes, roughly $600 billion of SpaceX's stock value vanished shortly after the company announced its acquisition of Cursor, the AI coding agent. The drop appears to reflect strong investor concern over the deal. Around the same time, reports surfaced offering context for Elon Musk's acquisition of the autonomous coding startup, citing the view that "China is on the verge of overtaking the United States in AI." It's a clear sign that the race to snap up AI coding tools is taking on growing geopolitical significance.

NYU Prof. Damodaran: "An AI Crash Would Be Even Harsher Than the Dot-Com Bubble"

Aswath Damodaran, professor of finance at NYU, has issued a warning about a potential collapse of the AI market. The crux of his argument is that, unlike in the dot-com era, the AI industry has built "massive physical infrastructure funded with borrowed money." In contrast to the software-centric bubble of that period, investment has swelled into heavy assets such as data centers and semiconductor fabrication facilities, which he argues would make the economic shock of any crash far more severe. He also notes that even if AI succeeds, its actual business model amounts to "replacing the work itself," leaving its impact on society opaque.

Anthropic in the Crosshairs Over AI Export Controls?

According to a Financial Times report (making the rounds on Hacker News), questions are being raised about whether Anthropic — through its own statements and lobbying — has ended up contributing to a push toward tighter export controls on AI technology. The possibility that the direction of regulation could work in favor of specific companies has sparked debate within the industry. It remains speculative at this point, but it has become one reason AI companies' involvement in regulation is drawing scrutiny.

GLM 5.2 Delivers High-Quality SVG Generation Even at an Extremely Low Quantization (IQ2_M)

The SVG image generation quality of GLM 5.2 running at IQ2_M (an extremely low quantization) is drawing attention in the Reddit LocalLLaMA community. The poster tested it on an RTX 5090 + RTX 3090 setup and reported: "I wasn't expecting much given the low quantization, but it generated a surprisingly high-quality SVG image of a pelican." While the tokens-per-second rate is low, they rate the model's underlying potential as very high. It's a fascinating case demonstrating the growing practicality of open-source models running in local environments.

AI Agent Security Threat "AutoJack" — and the Arrival of Defensive Tools

An attack technique called "AutoJack" was introduced on Qiita. It hijacks AI agents that trust localhost with nothing more than a single web page. In a proof-of-concept, merely instructing an agent to "take a look at this page" launched calc.exe through the agent's host process. Reported by Microsoft security researchers, it shows that the local trust model of AI agents can become a new attack surface.

At the same time, tools designed to make AI agents safer are emerging. "Lelu," a hot topic on Hacker News, is an authorization engine that detects manipulated AI agents. "OverReach" is a tool that audits whether AI agents have made changes beyond what their prompts instructed. As AI agents proliferate, practical countermeasures are taking shape on both the security and governance fronts.

---