Google Moves Gemini Into the Classified Zone
Google is in talks to deploy Gemini in classified Pentagon environments for "all lawful uses," with only contractual guardrails against autonomous weapons and mass surveillance. The shift raises a pointed question: when the contractor and the regulator are the same entity, how much does a contract restriction actually hold?
Read full story →The Model You Can Use, and the Model You Can't
Anthropic's new Claude Opus 4.7 is its strongest publicly available model, but the company has explicitly stated it is less capable than Claude Mythos Preview, which found thousands of zero-day vulnerabilities across every major OS and browser. Opus 4.7's cybersecurity limitations are not architectural; they are deliberate choices.
Read full story →The Race to Own the Developer's Desktop
OpenAI has given Codex the ability to run multiple agents in the background on Mac, working across apps while you keep coding. Coming weeks after Anthropic's Claude Code moved into desktop control, the update makes the competition for the developer's machine explicit.
Read full story →Central Banks Are Treating AI as the Next Systemic Risk
The Bank of England has formally added AI to its annual financial system stress tests, a structural shift triggered by Anthropic's Mythos model and months of escalating warnings from BOE Governor Andrew Bailey. The question is whether the stress test framework is even the right tool for a risk this different in shape.
Read full story →The AI Industry Is Buying the Midterms
Andreessen Horowitz's latest $25 million brings the "Leading the Future" super PAC above $51 million, with the AI industry's total 2026 election spending now at $140 million. For the first time, AI has become explicitly contentious in individual House and Senate races, and the counterweights operating at comparable political scale are essentially absent.
Read full story →Who Gets to Make the Rules?
The AI industry's $140 million midterm war chest is not a scandal; it is the market doing what markets do. What it actually closes off is the regulatory window: the last democratic mechanism with real teeth for shaping how AI gets deployed. Once the political framing sets, adjusting it becomes structurally hard. And p(sustainable) requires that window to stay open long enough for societies to make informed choices.
Read opinion →