Newslit News

SpaceX just bought the AI coding tool that competes with Claude Code. Cursor, the coding assistant developers swear by, signed a deal with SpaceX at a $10B initial payment, with a potential $60B valuation on the table. The kicker: Cursor will now run on xAI's Colossus, a supercomputer equivalent to a million H100s. SpaceX absorbed xAI in February. Their goal, stated plainly: "create the world's best coding and knowledge work AI."

What does "reasoning" actually change for image AI? Today's Newslit Daily breaks it down. newslit.substack.com/p/chatgpt-thin…

OpenAI just gave image generation a thinking brain. The new gpt-image-2 model doesn't just generate. It reasons: it plans the composition, counts objects, checks constraints, and then renders. The result is 2K resolution, aspect ratios from 3:1 to 1:3, and up to 8 coherent images per prompt. Oh, and DALL-E 2 and 3 are retired in May.

OpenAI shipped a new trick for Codex on Mac called Chronicle. Instead of you pasting context into every prompt, the coding agent looks at your recent screen activity and figures out what "this error" or "that document" means. Over time it learns your workflows, tools, and recurring projects. Screenshots and memories stay local on the Mac. You can inspect, edit, or turn it off from the menu bar. The catch: OpenAI says it burns through rate limits fast, and it is Pro-only for now.

GitHub Copilot is about to get more expensive, and Microsoft is pausing new signups while it figures out the pricing. Leaked internal documents from Ed Zitron show the company is moving from request-based to token-based billing, pulling Anthropic Opus 4.5 and 4.6 from the $10 Pro tier, and tightening rate limits across every plan. Why now? Weekly operating costs have nearly doubled since January. This is the subsidy era of AI coding tools ending in real time, and Copilot users are the first to feel it.

Full breakdown in today's Newslit Daily. newslit.substack.com/p/ternus-takes…

After fifteen years, Tim Cook is handing over the keys. Apple just named John Ternus as its next CEO, effective September 1. He has been at Apple since 2001 and has quietly run hardware engineering for years, shipping Vision Pro, Apple silicon, and most of the iPhone lineup you own. Cook stays on as executive chairman and will shadow the transition through summer. The board approved it unanimously. This is the first real CEO change at Apple since Steve Jobs handed off to Cook in 2011. What happens next shapes the next decade of the most valuable company on earth.

Most tech companies avoid the culture wars. Palantir just pinned its to the homepage. A new 22 point summary of Alex Karp's book argues Silicon Valley owes America a moral debt, defends military AI outright, and calls some cultures "regressive and harmful." Bellingcat's Eliot Higgins calls it a sales pitch for ICE contracts dressed up as philosophy.

Two months ago the White House told federal agencies to stop buying Anthropic. Today, reporting says the NSA is already running the company's newest model, Mythos Preview, against computer security problems. About forty organizations have access. The meeting between Dario Amodei and Susie Wiles last week starts to make more sense.

The full rundown, plus what Vercel customers should rotate today, is in this morning's Newslit Daily. newslit.substack.com/p/vercel-hacke…

Your deployment platform may have leaked more than you think. Vercel confirms attackers slipped into its internal systems via a compromised Google Workspace account at a small AI vendor. From there they pulled source code, API tokens, and the "non sensitive" environment variables nobody thought to encrypt. ShinyHunters is now on the forums advertising 580 employee records for sale.

Brussels rolled out its shiny new age verification app this week with the usual fanfare about privacy and open source, and within hours researchers had ripped it apart. One consultant says he broke it in under two minutes. The biometric lock can be bypassed. Sensitive data sits unprotected on the phone. And because of how the logic works, once you verify as an adult, anyone else holding your device can pass as one too. Turns out 400+ experts had already asked the EU to pause the rollout. They were ignored.

OpenAI quietly turned Codex into something different this week. The desktop app on Mac can now see what's on your screen and actually use it, clicking around, typing into apps, running multiple agents at once without getting in your way. It also picked up an in-app browser for frontend work, native image generation, and a memory that carries across sessions so you stop repeating yourself. Plus 90 new plugins. It's starting to look less like a coding helper and more like a full workstation copilot.

Full breakdown (plus today's other stories) in today's Newslit Daily. newslit.substack.com/p/google-pairs…

Google just quietly rewired how search works on Chrome desktop, and it's the kind of change you don't fully get until you see it. Now when you tap a link inside AI Mode, the page pops up right next to the chat. Ask follow-ups, compare details, jump between tabs without losing your train of thought. They even added a "plus" menu so your open tabs, notes, and images can feed straight into the search. Small update, bigger implications for how we browse.

Google just rolled out a new Gemini text-to-speech model, and the big idea is control. Developers can now steer things like tone, pacing, accent, and even multi-speaker dialogue with simple prompts, which makes AI voice tools feel a lot more flexible. It also gives Google a stronger play in a fast-growing corner of generative AI that is getting more competitive by the week.

Check out this story, and others, in the full newsletter. newslit.substack.com/p/claude-id-ba…

A lot of people moved to Claude because it felt like the more privacy-minded AI option. Now Anthropic is asking some users for a government ID and even a live selfie to unlock certain features or pass integrity checks.

Claude is reworking its desktop coding app around a simple idea: developers are no longer working with one agent at a time. The new setup is built for juggling multiple sessions, reviewing diffs, running tests, and editing files without bouncing between tools.

Microsoft just rolled out a cheaper, faster image model, which sounds like a routine product update until you look at what it says about the bigger strategy. This is not only about generating pictures. It is about building more of the AI stack in-house and making those tools practical for high-volume business use.