Teens are upset that they can’t speak to Character.AI chatbots anymore

The platform has restricted access for those under 18, as concerns around the tech’s impact grow.

11/25/25 10:12AM

As far as execs at Character.AI are concerned, (role) playtime’s over for under-18s, as they start to ban teens from using the platform’s chatbots this week.

The chatbot company, which clocks about 184 million website visits around the world each month — and millions more app sessions — per data from Similarweb, announced at the end of last month that it would be rolling back access to its open-ended chat feature for minors. Users who are under 18 have been limited to daily two-hour open-ended chats since the late October notice, and access started ramping down for most of those in the age range yesterday.

Indeed, a selection of young “power users,” as The Wall Street Journal termed them in a piece on the anguish some under-18 users are experiencing at being separated from the chatbots, are being given a weeks-long grace period where they can still access hour-long open-ended chats with the characters they’ve been interacting with, to “help minimize disruption” to the teen users.

Clearly, AI-generated conversations with popular characters like Yor Forger, “a loving mom who’s definitely not an assassin,” and Itoshi Rin, who “only loves soccer... and you,” have a lot of (mostly younger) users in the site’s full thrall.

Character.ai demographic chart — Sherwood News

Though traffic-tracking site Similarweb doesn’t break out under-18s in its demographics data, the figures it does disclose already show that Character.AI tends to skew a lot younger than more general competitors like ChatGPT.

According to US-specific figures from Similarweb, of the 1.7 million monthly unique web visitors character.ai notched on average from August to October, some 52% were aged between 18 and 24, compared to a 26% share on chatgpt.com.

A lot to talk about

Character.AI’s younger users tend to be way more locked in, too: per the data, they visited 26 times each month on average and spent 18 minutes in a typical session using Character.AI — much higher than ChatGPT users, who visited the OpenAI chatbot 13 times on average across the month, spending around 6 minutes per visit.

Those sorts of numbers, paired with deaths linked to the platform and increasing concerns around the detrimental effects of chatbots on younger minds more generally, were likely factors in the move to pare back teen user access, which — at least according to Character.AI CEO Karandeep Anand — “wasn’t a very hard decision.”

Jon Keegan3h

Delays in permitting, power, and zoning cause first drop in data center construction since 2020

Despite incredible demand, the number of data centers under construction in North America fell for the first time since 2020, according to new research from CBRE.

Total data center capacity under construction dropped about 5.6% year on year from 6.35 megawatts in 2024 to 5.99 megawatts by the end of 2025.

What’s causing the delay? Slow permitting, constrained supply chains, and growing public engagement with how deals are approved at the local level. Labor constraints also were cited in the report; a tight supply of skilled workers will increase costs.

US Data Center Construction Fell Amid Permit and Power Delays

-13%📱

Rani Molla5h

Smartphone shipments are expected to decline 13% — the biggest drop ever — to 1.12 billion in 2026, according to new data from IDC, as the memory shortage drives up costs and prices for phones. The firm expects the average smartphone selling price to jump 14% to a record $523 this year.

The shortfall will mostly affect makers of lower-end smartphones, whose customers are more cost-conscious, while higher-end manufacturers like Samsung and Apple are likely to be more insulated from the pressure.

“The memory crisis will cause more than a temporary decline; it marks a structural reset of the entire market, fundamentally reshaping long‑term TAM (Total Addressable Market), the vendor landscape, and the product mix,” said Nabila Popal, senior research director with IDC’s Worldwide Quarterly Mobile Phone Tracker. “We expect consolidation as smaller players exit, and low-end vendors to face sharp shipment declines amid supply constraints and lower demand at higher price points.”

Jon Keegan8h

Google drops new Nano Banana

Google is hoping to recapture the viral boost it received when it released its Nano Banana image generation model. Nano Banana 2 arrives today, which Google has rolled into its Gemini app.

The new model promises more accurate text rendering and translation and “advanced world knowledge,” which “pulls from Gemini’s real-world knowledge base, and is powered by real-time information and images from web search to more accurately render specific subjects,” according to the company’s press release.

New creative controls let users keep groups of characters consistent across scenes, render images with higher resolution, and parse complex prompts.

The first version of Nano Banana became popular for making action figures out of users, and helped catapult the Gemini AI app to the top of the charts, bumping ChatGPT from its perch.

Nano Banana 2: Combining Pro capabilities with lightning-fast speed

New creative controls let users keep groups of characters consistent across scenes, render images with higher resolution, and parse complex prompts.

The first version of Nano Banana became popular for making action figures out of users, and helped catapult the Gemini AI app to the top of the charts, bumping ChatGPT from its perch.

Rani Molla9h

Tesla’s ride-hailing service is looking a lot more like Uber’s than Waymo’s

Despite numerous promises about amassing a giant network of driverless cars, so far it seems like Tesla’s Robotaxis are a lot more similar to Uber’s plain old ride-hailing service than Waymo’s expanding autonomous fleet.

In California, where Tesla has its largest ride-hailing service, the company has taken no formal steps to gain approval for a truly driverless car service, according to Reuters. Throughout 2025, Tesla failed to log a single mile of autonomous test driving on state roads, and has not applied for the necessary permits to test or deploy vehicles without a human present. Currently, Tesla holds only a basic permit that requires a human safety monitor to remain in the driver’s seat at all times.

Currently, Tesla’s California Robotaxi service consists of roughly 300 Teslas operated by human drivers using the company’s supervised Full Self-Driving tech. In Austin, where the company has about 45 vehicles, Tesla made a big show earlier this year of announcing it was removing the safety monitors sitting in the front seats during rides. However, to date, only a handful of those vehicles have been reported to be actually operating without a safety monitor onboard.

In other words, it’s performing a service more akin to a tech-heavy Uber ride than the one operated by Alphabet subsidiary Waymo, which earlier this week announced it now has driverless rides available to the public in 10 markets. Even Uber is trying to put space between itself and the old driver-having Ubers of yore: this week its autonomous software partner said the company plans to launch a driverless service in London this year, with plans for 10 markets.

During its earnings report last month, Tesla said it planned to offer Robotaxi service in a half dozen new cities in the first half of this year, including Phoenix, Miami, and Las Vegas. Judging by Tesla’s progress so far, it’s likely those services will also feature a human in the front seat.