Greg Brockman offers his account of the negotiations and falling-out that led Elon Musk to exit OpenAI's founding team. The testimony is surfacing as trial evidence, giving rare visibility into OpenAI's early governance and power dynamics. Primarily historical context, but relevant for understanding the structural tensions now playing out in court.
Courtroom color piece covering the first week of the Musk v. Altman trial in Oakland, describing the atmosphere as two of AI's most powerful figures begin their legal face-off. The trial's outcome could set precedents around nonprofit-to-for-profit conversion in AI and donor intent law. Low HN signal suggests limited technical interest, but governance implications are real.
Musk testified that OpenAI's leadership deceived him about the company's mission, while under questioning he admitted xAI uses distillation from OpenAI's models — a significant IP disclosure with potential legal and competitive fallout. The distillation admission is the most technically and commercially consequential detail to emerge from week one. If substantiated, it raises questions about the legality and prevalence of model distillation across the industry.
Andon Labs is running an AI-operated cafe in Stockholm, following their earlier AI-run retail store experiment in San Francisco — using AI agents to handle inventory, ordering, and operations in a real physical business. Early results surface interesting failure modes (e.g., over-ordering 120 eggs) alongside genuine operational insights about where AI agents succeed and break down in physical-world logistics. This is one of the most concrete public experiments in AI-operated brick-and-mortar business.
Moonshot AI, maker of the Kimi model, has raised $2B at a $20B valuation with $200M in annualized recurring revenue driven by subscriptions and API usage — signaling strong product-market fit for Chinese frontier AI. The raise reinforces that the Chinese AI ecosystem is producing well-capitalized, revenue-generating competitors, not just research labs. Combined with DeepSeek's pending $45B valuation round, this marks a structural shift in the global AI competitive landscape.
DeepSeek, the Chinese AI lab that shocked the industry in early 2025 with frontier-quality models trained at a fraction of US costs, is reportedly targeting a $45B valuation in its first external funding round. This would make it one of the most valuable AI companies globally despite being a relative newcomer to external capital markets. The valuation reflects both its technical credibility and the strategic value Western investors and partners assign to cost-efficient frontier AI.
Daemon Tools, a widely installed disk image utility on Windows, was compromised in a supply-chain attack lasting roughly one month, delivering backdoor malware to users who downloaded or updated the software. This follows a pattern of attackers targeting developer and power-user tools to maximize infection surface. Any machine with Daemon Tools installed during the compromise window should be treated as potentially breached.
A critical Linux vulnerability dubbed CopyFail is being described as the most severe Linux threat in years, with confirmed impact on multi-tenant servers, CI/CD pipelines, and Kubernetes environments. The breadth of affected surfaces — spanning cloud infrastructure, container orchestration, and automated build systems — makes this unusually dangerous for AI infrastructure operators. Patching is urgent; unpatched systems in shared compute environments are at high risk of lateral movement.
Simon Willison observes that vibe coding — exploratory, low-oversight AI-assisted coding — is converging with agentic engineering workflows in his own practice, which he finds unsettling. The concern is that as agentic systems take on more autonomous execution, the low-scrutiny habits of vibe coding create compounding risk when agents act on flawed code. With an HN score of 1468, this is clearly resonating broadly with the builder community.
Google DeepMind is acquiring a minority stake in CCP Games, the studio behind EVE Online, to use the complex space MMO as a testing environment for AI models. EVE Online's persistent, multi-agent, economically complex environment makes it a uniquely rich simulation for testing strategic reasoning, emergent behavior, and multi-agent coordination at scale. This follows DeepMind's historical pattern of using games (Go, StarCraft, Atari) as research proving grounds before productizing capabilities.
That's today's briefing.
Get it in your inbox every morning — free.
Help us improve AI in News
Got a suggestion, bug report, or question?