The week handed us a near-perfect dialectic. On one side, the Musk vs. OpenAI trial opened, two parties arguing in court about what AI was supposed to be: a nonprofit mission, a commercial empire, or some impossible hybrid. On the other side, Google quietly signed a classified deal with the Pentagon allowing use of its AI models for any lawful government purpose. The trial is about the past. The Pentagon deal is the future. They describe the same arc.
The 'Lawful' Modifier Is Doing Enormous Work
The phrase any lawful government purpose is a legal construction that sounds like a constraint but functions as a permission structure. Lawful shifts over time. Lawful expands with executive authority. Lawful in a classified context is not reviewed by the same public processes that lawful in a consumer context implies. The Atlantic's piece on OpenAI imitating Anthropic frames the AI safety competition as a branding war, with OpenAI racing to seem more principled than it is. But the week's actual data point is Google, not OpenAI, signing the most expansive AI-to-government handshake yet reported.
The Script Kiddies Knew First
The Verge's deep read on AI-enabled cyberattacks describes how AI has democratized offensive hacking capability in ways that outpace defensive infrastructure. Replit's CEO, flagged in this week's investor signals, has already predicted every company becomes a cybersecurity company by necessity. The arXiv paper on AGI forecasting methodologies argues that scenario analysis for AI risk is methodologically immature compared to the pace of deployment. The Musk trial is asking what OpenAI promised in 2015. TurboFund's list of active AI VCs in 2026 reflects a market that has entirely moved past these questions. The capital has decided. The courts are still catching up.