Roundup: everyday AI alarm and adoption stories

The immediate trigger traces back to a mundane event: an online auction of the original 2003 'Cat in the Hat' film costume put the character back into public conversation, and users combined that renewed attention with consumer-grade generative AI to place a menacing, oversized version of the character into fabricated 'doorbell camera' footage of familiar residential streets. The trick worked because Ring-style security footage carries an implicit stamp of authenticity - grainy, framed like a real alert - and needs only a text prompt to fabricate. Police statements had to explicitly rule out a literal threat: "Fortunately, we can rule out the issue, which is essentially an AI prompt that places the character into images/videos of familiar places" ^[2].

What made this more than a one-town joke was how fast and far it traveled: sightings piled up across Limerick, Cork, Cahir, Tipperary, Wexford and Dublin in Ireland, and Sheffield and Glasgow in the UK, pulling in Gardai in Wexford, South Yorkshire Police, and Police Scotland, each forced to issue its own public correction ^[1]. One force summed up the absurdity of having to respond at all: "In an unusual turn of events we are hearing a lot about an illusive Dr Seuss character doing the rounds online" ^[1]. YouTube commentary tracked the same mechanic from the outside - creators noted the trend borrowed horror-game visual grammar and spread partly through crowd-sourced fictional backstories that got remixed and re-uploaded, turning a single meme into an evolving, self-reinforcing hoax rather than a one-off fake. On Reddit, reaction skewed skeptical and mocking rather than frightened, with commenters drawing the same comparison police themselves were implicitly fighting: this looked like a rerun of the 2016 creepy-clown panic, dressed up in AI production values good enough to fool people who don't stop to check.

While the Cat in the Hat hoax was forcing multiple police departments into damage control, Apple was quietly doing the opposite: pushing a genuinely new AI feature into a mainstream app with no announcement at all. The company began testing a virtual shopping assistant inside the Apple Store app on iOS, limited to a small early-preview user pool, that can field questions about products, existing orders, trade-in values, financing, and carrier offers, and even compare devices like different iPad models ^[3]. The assistant introduces itself plainly - "I'm your virtual shopping assistant. Ask me about Apple products, existing orders, connecting with a specialist, and more" ^[4]- with none of the fanfare that typically accompanies a flagship AI feature.

The rollout comes with its own quiet disclaimers rather than a marketing push: Apple states the assistant "is powered by artificial intelligence and may give answers that contain inaccuracies," and that Apple "may use and retain a transcript to ensure quality and system performance, and to monitor for fraud" ^[3]. That combination - phased access, muted framing, upfront accuracy caveats - reads as a deliberate strategy to let a company iterate on an AI feature's rough edges before it becomes a headline. It is also why the feature has barely registered outside tech press: on Reddit, the dominant reaction was closer to a shrug than excitement, with one top comment dismissing it as the same kind of chatbot that has lived on Apple's store pages for years, and another user noting they had already had access for weeks before it made news. That muted reception is itself the point: Apple's approach only works because it deliberately avoids inviting the kind of scrutiny - or virality - that turned the Cat in the Hat hoax into a cross-border incident.

A decade before this week's stories, Geoffrey Hinton delivered arguably the most quoted AI-replacement prediction in medicine: in 2016 he said radiology would be automated away within five to ten years, telling radiologists they were 'like the coyote that's already over the edge of the cliff but hasn't yet looked down' ^[5]. That prediction has aged badly by every employment metric available. Average radiologist compensation reached $571,000 in 2025, up 9% year over year; the active radiologist population grew roughly 10% over the past decade; and as of May there were 7,469 active radiology job openings, with 1,470 unfilled for more than 60 days ^[5]. Darden School of Business economist Christoph Herpfer put it bluntly: 'We actually have a huge shortage of radiologists. So the exact opposite of this prediction has happened' ^[5]. Rad AI cofounder Dr. Jeff Chang goes further, dismissing the premise itself: 'That entire concept didn't really make sense to begin with' ^[5].

But dismissing Hinton's timeline entirely would be its own overcorrection. A Stanford-affiliated task-based modeling study posted to medRxiv in December 2025 estimates that automation of report drafting and study delegation for radiography and mammography could still cut radiologist hours worked by 14% to 49%, with a central estimate around 33%, over the next five years - even while concluding that 'radiologist job loss is unlikely for the foreseeable future,' because a largely static workforce is absorbing steadily rising imaging volumes ^[6]. That is a materially different claim than Hinton's: not that radiologists disappear, but that the same number of radiologists get more automated leverage per hour, with demand growth soaking up the slack rather than headcount shrinking. Community discussion on Reddit's radiology forums frames this as an efficiency-versus-demand tension rather than a replacement story, describing current tools like automated fracture-detection systems as still error-prone enough in daily use that full automation remains a distant proposition. Interventional radiologist Dr. Tonie Reincke's rebuttal to Hinton captures the piece no model can touch: 'A computer can't hold a patient's hand when they're crying' ^[5].

Line up all three stories and a single pattern emerges: the same underlying technology is simultaneously the subject of genuine public fear and the quiet infrastructure being folded into daily life without complaint. A hoax exploiting cheap, convincing generative video forced multiple police departments across two countries to spend real resources debunking a fictional threat ^[1]. In the same week, Apple slipped a genuinely new AI assistant into one of the most-used commerce apps on the planet with barely a ripple of public reaction ^[3]. And in radiology, a profession that has spent a decade bracing for AI-driven obsolescence is instead seeing demand and pay climb, even as credible modeling shows AI is already automating a meaningful share of the actual workload ^[6].

Survey data helps explain why both reactions are true at once rather than contradictory: nearly two-thirds of Americans say they have already adopted AI into their daily routines, yet 56% report at least some anxiety about AI's societal implications, and only 42% feel excited about it ^[7]. That gap - high adoption, low enthusiasm, persistent unease - is exactly the texture of this week's news. People are using AI-powered assistants and benefiting from AI-augmented workflows even as they remain primed to believe the worst about what AI-generated content can do to them, which is precisely why a fake doorbell-camera video could travel faster and further than Apple's actual AI product launch. The lesson for anyone tracking AI adoption isn't that the public is inconsistent - it's that trust and usage are decoupling, and the fear stories will keep outrunning the adoption stories in attention even as the adoption keeps quietly winning in practice.