China would not let Meta keep Manus. Users have until 24 August to save their work
THE SO WHAT
Regulators just forced a cross-border AI acquisition to unwind — and users are losing data in the process. If your workflows depend on AI startups with complex ownership or China exposure, add “what happens to my data if this deal is reversed” to your vendor diligence.
READ THE SOURCE
MORE FROM THE WIRE
Applied AISkan AI raises $63m to watch how office staff actually work, then build agents that copy them
The scarce asset isn’t the model, it’s the behavioral trace of how large enterprises actually get work done—Skan is raising $63M to own that corpus and then automate against it. If you run a process-heavy org, assume your keystrokes and clicks are now training data and decide whether you want that dataset owned by a vendor or captured in-house.
Applied AINvidia-Backed Startup CodeRabbit Valued at $1.5 Billion in Round
A $143M round at a $1.5B valuation for AI-powered code review says the market believes verification, not generation, is the new bottleneck. Engineering leaders should budget more for tooling that inspects AI-written code than for yet another assistant that writes it.
Applied AI‘New models will mark AI-generated content from day one’: Claude will now hide an invisible watermark inside ordinary words — here’s how that’s even possible, and how EU rules could push OpenAI and Google to follow suit
Anthropic baking invisible watermarks into all new Claude text is a preemptive compliance move—if EU rules harden, provenance becomes table stakes. Enterprises deploying generative content at scale should assume regulators and platforms will expect detectable signatures and start logging which outputs are marked and where they flow.
Applied AIMusk says SpaceX AI will out-earn rockets by September, and be trained on staff
Claiming SpaceX AI will out-earn rockets within a month and training Grok on internal data reframes the company as a vertically integrated AI + infra play, not just launch. For operators, the takeaway is blunt: your proprietary workflows and telemetry are now potential P&L lines if you can safely turn them into model training fuel.