OpenAI Will Delay GPT-5.6 After Trump Administration Request
The Trump White House — which spent considerable energy over the past year framing AI safety concerns as regulatory capture by incumbents — has now asked OpenAI to stagger the release of GPT-5.6 over, you guessed it, safety concerns. Sam Altman told employees the model will launch in limited preview to a small group of partners rather than the general public. I’m not saying the administration is wrong to pump the brakes, but the ideological whiplash here is genuinely something to behold.
Anthropic Thinks Its Own Success Is Key to Making AI Safe
Anthropic’s argument, stripped to its bones, is: the powerful AI is coming regardless, so it’s better if we control it. Critics are pointing out that this logic could justify virtually any accumulation of power by any sufficiently self-confident actor. Wired’s piece lands at an uncomfortable moment — Anthropic just filed for its IPO, meaning the company whose entire brand is safety-first is now also a company with fiduciary obligations to shareholders. Hold those two thoughts together and see how comfortable you feel.
Anthropic’s Claude Is Winning Over Paid Consumers, a Market Owned by ChatGPT
In the market that actually matters — people willing to open their wallets — Claude is eating into ChatGPT’s lead. This is the horse race that will determine whether Anthropic’s IPO is a triumph or an embarrassment, and right now the numbers are trending in their favor. If you’re wondering why OpenAI has been on an absolute product velocity tear lately, this data is probably a good chunk of the answer.
The $27 Million AI Proxy War Over Alex Bores Ends in a Draw
The most expensive local election in AI history ended with the pro-AI super PAC’s candidate — Alex Bores — narrowly losing the Democratic primary for New York’s 12th district. Twenty-seven million dollars, and the net result is that the AI industry may have actually galvanized opposition to itself in one of the most politically engaged districts in the country. Great investment, guys.
British Police Built a Sprawling Crime-Prediction Machine. Some Results Couldn’t Be Trusted
Wired’s investigation into a UK regional police force’s predictive analytics experiment is the kind of story that should be required reading for every government official currently rubber-stamping AI procurement. The system was deployed, the results were sometimes untrustworthy, and the people making decisions based on those results often didn’t know that. The gap between “AI-powered” and “AI-reliable” remains vast, and in law enforcement, that gap has real human consequences.
Ford Had to Hire Back Former Engineers to Fix Mistakes Made by Its Automated Systems
Ford just topped JD Power’s initial quality rankings among mainstream automakers — and the victory lap includes a quietly stunning admission: their automated production and design systems made errors that required rehiring the experienced engineers they’d previously let go. This is a masterclass in what happens when you automate before you fully understand what the humans were actually doing. The good news is Ford learned the lesson. The bad news is how many other manufacturers haven’t yet.
Notion Killing Email App Since Most Users Use AI Agents Instead
Notion is shutting down its email app — acquired via Skiff — because users have migrated to AI agents to manage their inboxes instead. The company says it’s “going all in on agents.” This is a small story with big implications: we’re starting to see the first wave of products that were built for humans being quietly retired because AI agents have taken over their function. The app-as-interface era may be winding down faster than most people realize.
IBM Has Unveiled Chip Technology That Could Help Extend Moore’s Law Another Decade
IBM has built a prototype chip with roughly 100 billion transistors packed into a fingernail-sized area — twice the density of their previous best. If this scales out of the lab, it’s significant not just for AI compute but for every device that runs silicon. Most of the AI infrastructure conversation has been about software and data centers; it’s worth remembering that raw chip physics still has room to run.
Bottom Line
The week ends with an odd harmony: the government wants AI slowed down, Anthropic wants to be the one who controls it, British police already deployed it before it was ready, and Ford just finished cleaning up after trusting it too much — which means the central tension in AI right now isn’t capability, it’s accountability.