August 13, 2026
TechCrunch reported that some Claude users are furious that Anthropic's new watermarking system is exposing their AI use at work and in school. The complaints are telling: these aren't people worried about privacy in the abstract, they're people who got caught doing something they were told not to do. That distinction matters. A watermark that quietly tags AI-generated text isn't a betrayal of trust, it's a feature working as designed. If anything, this is the most useful thing Anthropic has shipped for the employers and instructors who've spent two years trying to figure out who's actually doing the work. My take: the backlash says more about how normalized quiet AI use has become in workplaces than it does about Anthropic overstepping. Companies that have already put governance in place -- clear policies on where AI is allowed, and where a human still has to sign off -- aren't the ones complaining today. That's a good argument for building role-based permissions and clear-use policies before a watermark forces the conversation for you.
Cognition is reportedly in talks to raise at a $40 billion valuation, according to TechCrunch, barely months after closing $1 billion at $26 billion. In the same news cycle, Lovable confirmed a $13.3 billion valuation on a new $400 million raise after hitting $500 million in annualized revenue, and Blacksmith's valuation jumped nearly tenfold in under a year on the back of AI-driven code testing demand. Individually, each of these is a strong growth story. Together, they describe a market where investors are betting that AI coding tools aren't just useful, they're the new default way software gets built. That's a real trend -- ViibeStack customers ask us constantly how AI-assisted builders compare to hand-coding, which is part of why we wrote about Lovable prototypes moving to production and how ViibeStack stacks up against Cursor. But valuations moving this fast, this often, in a category that's still figuring out retention and margins, should make any business buyer a little skeptical. A $40 billion price tag assumes Cognition's growth curve holds for years, not quarters. If it doesn't, the correction won't just hit Cognition -- it recalibrates how every AI coding vendor gets valued and sold to enterprise buyers.
Amazon will train AI models on Twitch streamers' content by default unless they opt out, and Twitch's chief product officer said plainly that an opt-in model would get essentially zero participation. Credit for the candor, but it doesn't change the substance: this is the same playbook Meta, Google, and others have used, just said out loud instead of buried in a terms-of-service update. For businesses that generate any kind of user content -- video, chat logs, uploaded files -- this is a preview of the decision every platform vendor eventually faces. The safer path, and the one we think holds up better over time, is transparency about what's used for training and genuine control for the people producing the content, not a default that quietly assumes consent. It's the same principle behind how we think about data privacy and trust commitments generally: default-on data use might win short-term training data, but it costs long-term goodwill from the people whose content built the platform in the first place.
Which of these bothers you more: a tool that quietly flags when you used AI, or a platform that quietly assumes it can train on your content unless you say no?
Sources