Anthropic Cuts Deregulation Deal as GPT-5.6 Goes to Spooks

The Capture Is Bilateral Now

Anthropic, the lab that built its brand on being the worried one, is reportedly close to a deal with the Trump administration to strip restrictions from its top models. The framing is national security. The mechanism is deregulation. The company that spent years publishing scaling-risk papers is now negotiating the removal of the guardrails those papers justified, in exchange for a friendlier posture from Washington.

On the same day, OpenAI quietly rolled out GPT-5.6 as a limited preview for vetted US government users, with enhanced cybersecurity features baked in. Note the sequencing. The federal customer gets the new model first, hardened for their use case, before the rest of the market sees it. This is not a consumer product company anymore. It is a defense contractor with a chatbot side hustle.

Taken together, these two stories are the same story. The frontier labs are no longer lobbying the state from the outside. They are inside, with badges, negotiating terms. Whatever you thought AI governance was going to look like, it looks like procurement.

The Inflation Nobody Priced In

An IMF economist surfaced a thesis worth chewing on. The AI wealth boom, concentrated in equity holders of a handful of companies, is going to translate into consumer spending, which is going to push inflation higher. The wealth effect is real, the spenders are concentrated, and the productivity gains the bulls keep promising have not shown up in the price index yet.

This is the awkward middle phase. AI is rich enough to inflate asset prices and household balance sheets, but not yet productive enough to deflate the cost of actual goods and services. Central banks have not built a model for this. They will improvise, badly.

Meanwhile SpaceX, fresh off a $25 billion debt offering, watched its bonds slide to junk-rated yields within days. Even the marquee names are starting to find the edge of investor patience. The money is still flowing, but the terms are getting uglier, and the gap between private valuations and what bond buyers will actually pay is widening. Somewhere in that gap is a future writedown.

The Talent Drain Continues, the Distractions Multiply

Apple's Vision Pro chief, the executive in charge of the company's most expensive recent bet, just left for OpenAI. Apple keeps losing its AI and adjacent leaders to companies that pay in pre-IPO paper and promise to actually ship things people use. The Vision Pro was supposed to be the next platform. Instead it became the launchpad for its own leader's exit interview.

Zuckerberg, never one to miss a trend, is pushing Meta into prediction markets through a partnership with Polymarket inside a new app called Arena, aimed at the 18-34 demographic. The pitch is engagement. The product is gambling on news events, dressed up as information aggregation. If you thought the attention economy had hit its floor, Meta has located a basement.

The Atlantic ran a column this week arguing the Trump era resembles a cruel simulation. It is not a serious thesis, but it captured something real about the current information texture. When prediction markets, AI policy deals, government model previews, and bond market wobbles all land in the same news cycle, the brain reaches for the simulation metaphor because no single frame holds.

One Loss Worth Marking

Om Malik, founder of Gigaom, died at 59. He spent two decades shaping how the press covered Silicon Valley, which means he shaped, indirectly, how the Valley came to see itself. The blog era he helped build is gone, replaced by newsletters, podcasts, and whatever it is that X has become. The reporters working the AI beat today are still standing on infrastructure he helped pour. Worth noting on a day when the industry he covered is busy negotiating with the state, shipping models to spies, and pulling executives out of every adjacent company it can reach.

The trajectory is set. The only question is the slope.