Big Tech keeps shipping AI features it has to yank days later

For two years, Big Tech had one reflex: put generative AI into every product, and ship it fast. This week that reflex hit a wall. Features were pulled within days. Platforms turned on the AI slop their own tools had helped create. And a European law arrived to make the retreat mandatory.

The pattern was the same each time. Ship first, watch the backlash, walk it back. What changed this week is how quickly the walking-back came, and how many companies did it at once.

The features that did not last the week

Google offered the sharpest example. It added its Nano Banana image generator to Google Earth on Thursday, then pulled it on Friday. People had used it to fake a nuclear plant in Iran, refugees at the US border and a bomb crater beside a hospital in Gaza. The tool lasted a single day.

Meta had already done the same in miniature. Weeks earlier it launched an Instagram feature that turned other people’s public photos into AI images. It removed the tool within a week, saying it had “missed the mark”. The reflex is to launch. The correction is to retreat.

Platforms turn on their own slop

The bigger shift is that platforms are now fighting the very content their AI tools make so easy. On Friday Snapchat said it would stop recommending fully AI-generated videos in its Spotlight feed, even when the creator labels them, TechCrunch reported.

It is not alone. LinkedIn added a “seems like AI slop” button and says it now blocks hundreds of thousands of automated comments a day. Researchers found 40% of its long-form posts were fully AI-written, Fortune reported. YouTube has cut payouts for template-made AI videos, and others are following.

The common thread is authenticity as a selling point. After flooding their own feeds, the platforms are trying to prove there are still humans on the other side.

Meta is running the other way

Not everyone is retreating. Even as it pulled the Instagram tool, Meta is pushing harder into AI content elsewhere. Mark Zuckerberg told investors its new Muse Image and Muse Video models would “dramatically expand the universe of content”. He also lifted the company’s spending plan above $130bn.

Investors were not sold. Meta’s shares fell about 11% over five days, a reminder that “more AI content” is no longer an automatic win with the market either.

The law that ends the argument

Then comes the part that is not optional. On 2 August the transparency rules of the EU AI Act become law. Makers of generative systems must mark their output as artificial, and anyone publishing a deepfake or AI-written text on a public-interest topic must label it visibly.

The reach is wide and the teeth are real. The duties apply to any provider whose output is used in the EU, wherever the company sits. Fines can reach €15m or 3% of worldwide turnover, the European Commission says. About 190 organisations, Google, Meta, OpenAI and Microsoft among them, have already signed a voluntary code to show they comply.

Put together, the week marks a turn. For two years the question inside these companies was whether they could add AI to a product. Now the question is whether they can defend it once it ships, and whether they must label it when they do. The ship-it-all reflex has met the backlash, and now the law.