Here’s what’s shaping AI right now - without the noise:

SpaceXAI (formerly xAI) and Cursor released Grok Bot, a beta app that turns Grok into a roster of always-on teammates, each running on its own dedicated computer, able to log into a person's apps and finish real work instead of just answering prompts.

Bots coordinate with each other in group chats, hand off specialist tasks, and learn new workflows by watching a person record their screen. A "Chief of Staff" Bot can direct the others. Work keeps running 24/7 with no laptop open.

The beta covers iPhone, Mac, Windows, and Linux, priced at $200 a month for individuals through Cursor Ultra and $120 per seat for teams. Everyone else goes on a waitlist.

The bigger signal: 👉 The AI-teammate-in-a-group-chat format keeps showing up across labs. SpaceXAI just built the version priced for the masses, if the price ever actually comes down.

AI is moving fast and most people are falling behind. 

The Rundown AI keeps you ahead of the curve. 

It's a free AI newsletter that keeps you up-to-date on the latest AI news, and teaches you how to apply it in just 5 minutes a day.

Plus, complete the quiz after signing up and they’ll recommend the best AI tools, guides, and courses — tailored to your needs.

Sign up to start learning.

Anthropic published a support page laying out how it now embeds invisible watermarks into text, code, and files that Claude generates, rolling out the EU AI Act's transparency rules across its entire user base rather than just Europe.

New Claude models carry a hidden watermark that survives copy-paste outside the platform. Files get the same C2PA provenance tag the industry already uses for AI-generated media. Anthropic says the mark only confirms content was "processed by Claude," not that Claude wrote every word of it.

Older models need retrofitting, models shipped after Aug. 2 already have it built in, and Anthropic is building its own detection tool. OpenAI, Google, and Meta will likely face the same requirement soon. xAI hasn't signed the EU AI Act.

The bigger signal: 👉 A watermark that survives copy-paste means AI-generated text carries its origin wherever it goes, and every major lab not named xAI is about to face the same choice.

Everyone's suddenly building always-on AI agents. Before you turn one loose, four things will save you from finding out the hard way what "24/7" actually costs.

Set a hard ceiling first, not after. Give your agent a daily token or dollar cap before it starts, not after you get a surprise bill. Most agent frameworks let you set this at the account or project level. If yours doesn't, wrap every agent call in a counter that kills the process once it crosses your number.

Treat everything the agent reads as untrusted. System instructions you wrote are high trust. Anything the agent pulls from an email, a webpage, or a file someone else touched is zero trust, no matter how official it looks. An agent that blindly follows instructions embedded in a scraped page is an agent waiting to be hijacked.

Log every autonomous action somewhere a human can actually read it. Keep a plain running file of what the agent did, when, and why it decided to do it, something you'd actually open. When things break at 3 AM, that log is the difference between a five-minute fix and a day of guessing.

Test your kill switch before you need it. Build the "stop everything now" command on day one. Then run it once a week just to confirm it still works.

Try this as a starting check-in prompt for any agent you're about to leave running unattended:

The bigger signal: 👉 Every lab is racing to sell you an agent that never stops working. The boring infrastructure that keeps it from working against you is still entirely on you to build.

Manus said it will go back to running as an independent company after Chinese regulators ordered Meta to unwind its $2 billion acquisition of the AI agent startup, ending a deal first announced in December.

China's National Development and Reform Commission issued the order in April, reportedly examining whether the deal violated foreign investment rules amid tighter export controls tied to the broader US-China AI competition.

As part of the split, Manus told some users they need to back up any data created on or after December 29, 2025, the date the Meta deal was first announced, to satisfy regional regulatory requirements.

The bigger signal: 👉 A $2 billion acquisition getting unwound by regulators is a preview of how much harder cross-border AI deals are about to get.

A few tools quietly worth exploring: