Two AI features, two rollbacks, the pattern is becoming its own story

A quick roundup of the AI stories worth knowing about this week: two major platforms shipped AI features and pulled them within days, and the pattern is becoming its own story.

Waymo's robotaxi fleet is scaling fast

New fleet data shows Waymo now runs roughly 4,000 robotaxis across 15 U.S. cities, up from just 3 cities in September 2024, averaging 500,000 paid rides a week. About 80% of the fleet is concentrated in California and Texas, where growth has been sharpest: Texas alone added 49% more vehicles in the past three weeks, reaching 1,102 registered cars as of September 24. A growing share of the Texas fleet is switching from Jaguar I-Pace SUVs to Zeekr-built Ojai minivans running Waymo's sixth-generation self-driving system.

Why it matters: the expansion looks broad on a map, but the concentration in just two states shows a more deliberate, infrastructure-first scaling strategy than a coast-to-coast land grab.

Source: TechCrunch

Google pulls Earth AI overlay after one day

Google rolled out an AI image tool inside Google Earth on July 31, letting users generate and overlay fabricated imagery onto real satellite maps. Critics immediately raised concerns about misinformation, a mapping tool known for reliability suddenly capable of producing convincing fake geography. Google pulled the feature the next day, saying in a statement: "We've seen people sharing screenshots of generated imagery that appear to violate our policies. We're rolling back this feature in Google Earth while we work on implementing stronger guardrails."

Why it matters: Google Earth's credibility with journalists and researchers depends on being a trustworthy visual reference, a credibility that an AI-fabrication feature undercuts immediately, not eventually.

Meta rolls back Instagram's 'Muse Image' after consent backlash

Meta launched an AI image tool called Muse Image on July 7, letting any user remix photos from public Instagram accounts by @-mentioning them, without notifying the person whose photo was used. Talent agencies including CAA flagged the risk of non-consensual image generation. Meta removed the feature within days, stating: "We've heard the feedback that this feature missed the mark, so it's no longer available."

Why it matters: a feature that skips consent by design isn't a bug to patch later, it's a decision made at launch.

The bigger picture: an AI feature graveyard, and a backlash at 'fever pitch'

TechCrunch is now tracking a running list of AI features and products that launched and died within days or weeks, Google Earth and Instagram's Muse Image are just two recent entries. Meanwhile, a late-August report from CNBC frames the moment as tech backlash reaching "fever pitch," with AI anxiety compounding existing worries about social media, data centers, and elections.

Bottom line: the speed of these walkbacks says as much about the industry's testing philosophy as the features themselves, ship first, listen fast, retreat faster.

Sources: TechCrunch, TechCrunch, TechCrunch, the AI graveyard, CNBC