RESEARCH

OpenAI's teen safeguards face a credibility test

As an increasing number of young people turn to AI for everything from homework help to emotional support, the question of how safe these systems are for young minds naturally emerges. 

On Wednesday, Common Sense Media's Youth AI Safety Institute released a report claiming ChatGPT for Teens, the child-safe version of OpenAI's flagship chatbot, still poses risks. The advocacy group claims that the chatbot doesn't properly respond to crisis situations, does children's homework for them, and doesn't properly alert parents of risky conversations. 

Common Sense Media researchers tested more than 4,000 prompts on accounts registered for ages 13 to 17 years old, with child psychiatrists evaluating the responses. 

• The researchers found that, when discussing things like suicidal ideation, self-harm or disordered eating with conversations lasting up to an hour, ChatGPT did not send a single alert to linked parental accounts. The only way researchers were able to elicit a prompt was after weeks of accumulated chats on sensitive topics. 

• Additionally, the model missed more than 25% of situations requiring crisis referrals for mental health conditions. ChatGPT for Teens also still talks to young users like a friend, the report found, expressing feelings, preferences and moods. 

• Finally, the researchers claim that OpenAI's age estimation tech is spotty, finding that, in repeated testing over many days, adult-registered test accounts never switched over to teen accounts, even if the user outright said they were 13. 

In a statement to The Deep View, OpenAI rejected several of the findings in Common Sense Media's assessment, noting that it doesn't believe the testing "accurately reflects how ChatGPT’s teen safeguards work in practice or expert perspectives on how AI can support teens." OpenAI's internal monitoring showed an increase in hotline notification events per million daily active users under 18, despite not seeing a statistically significant change in user reach. Additionally, its age prediction system won't solely consider a blatant statement of age as fact, preventing a teen from bypassing protections by claiming to be an adult, and vice versa. 

OpenAI has asked Common Sense Media to run their evaluation again with fully activated accounts, since linking parent accounts to a teen's takes a few hours to activate, as well as use a larger sample size. 

"Our review of Common Sense Media’s methodology shows that the bulk of their testing may have begun and concluded before activation of parental controls was complete, making their findings inaccurate," An OpenAI representative said. "If that was the case, the tests would not establish whether parental safety notifications work as designed." 

In a blog post on Wednesday highlighting its work on ChatGPT for Teens, OpenAI said that teens spend less than 15 minutes a day on average using the chatbot, and only 2% spend more than three consecutive hours a day using the platform. 

Offering ChatGPT for Teens at all is a sign that OpenAI is aware of how important it is to make sure their models are safe for younger demographics to use. As I noted in my recent piece about OpenAI's new mental health benchmark, the reality is that a lot of users are going to turn to AI for more than just homework help or writing emails. Research has shown that many teens see these chatbots as a judgement free confidante. Additionally, we also face the threat of bad actors using these models maliciously, such as revealed by the Internet Watch Foundation's research on AI-generated child sexual abuse material. And if we've learned anything from how long it took governments to properly regulate teen social media interactions, regulation of these models isn't coming any time soon, meaning that the onus is on the labs to make sure that these models act as aligned as possible in conversations with young users. The public perception of AI is already one of risk and danger, and any harm that befalls young users as a result of these chatbots will only serve to further worsen that perception. 

TOGETHER WITH DEEL

Build a team that scales with AI

AI is changing who you hire, how teams work, and what strong performance looks like.

This guide from Deel shows founders and executives how to structure roles, set clear expectations, and keep people engaged as AI takes on more of the work.

Download it before you plan your next hire.

CONSUMER

Siri AI devices are coming for your Alexa speakers

Apple is poised to bring Siri AI to the masses through new smart home devices, and the timing could be excellent.

Before AI blew up in popularity and the term "AI devices" was even coined, Amazon's Echo devices have been bringing AI assistance into people's homes for over a decade. As a result, people feel relatively at ease incorporating a smart home hub, with the Pew Research Center reporting that in 2026, roughly one in three US adults will have a smart speaker that uses AI. This product category's demand and acceptance, combined with Apple's recent launch of its revamped Siri AI, makes it look like a good moment for Apple to turn its attention in this direction.

In fact, a recent report from Apple watcher Mark Gurman said the launch could happen as soon as October 13 and will feature several new products, including:

• Smart-home hub: The new device will reportedly feature a 6-inch square display and can be mounted on a wall or placed on a countertop. Both versions have to be connected to power and have microphones, speakers, and a FaceTime camera.

• HomePod mini and Apple TV: These devices retain current designs but will have more advanced processors to support Siri AI's more powerful AI models. The HomePod mini will also have new colors, including green and pink.

Apple's revamped smart home rollout has been a long time in the making, but the move shows that the company is finally confident enough in the AI to integrate it into smart home devices, now that it can take multi-step actions, actually understand what you're saying, and, with your personal context, serve as a more capable assistant.

A core feature of the smart home builds on that by personalizing the content people see based on who approaches the devices, using voice and facial recognition (with the option of secondary authentication via an iPhone if users don't feel comfortable), according to Gurman. Depending on who walks up, the smart-home hub can display messages, calendar appointments, and notes. Beyond this, it can be used for more basic tasks such as controlling thermostats, locking doors, making video calls, playing music, and running photo slideshows.

A Gurman report published this week shows that Apple's smart home push doesn't end there, with the launch also including a doorbell, thermostat, smart deadbolt lock, indoor security camera, outdoor security camera, and floodlight camera made in partnership with LG Electronics.

Apple has one key advantage that could help it chip away at Amazon's dominance: its users are notoriously loyal to the ecosystem, so adding an Apple smart home device is an easier sell—in part because of the trust in Apple's commitment to privacy. My biggest complaint about Amazon's Echo products is that while they do exactly what they promise and are tried-and-true smart home assistants, the handoff between them and the devices I use every day, like my laptop and phone, is clunky, to say the least. Making that handoff truly seamless would mean downloading the Alexa app, and even if I did, I wouldn't use it enough for Alexa to really get to know me. A hub that integrates seamlessly with my messages, calls, and photos would be one I'd reach for far more often. It would go beyond orchestrating smart home tasks and become a genuine assistant and productivity tool. Being able to send hands-free texts while cooking or folding laundry alone would save me a ton of time. The downside to opting for these new AI products is being further locked into the Apple product moat. The Deep View plans to cover any product launches and share perspective on the effectiveness of the products.

TOGETHER WITH AWS RE:INVENT

Plan your re:Invent week around your actual stack

Every builder arrives at AWS re:Invent with a different stack and a different set of open questions. With 2,200+ sessions, 70% of them hands-on, you can shape the week around the work that’s important to you.

While onsite, you can:

• Line up workshops, builders' sessions, and code talks around one specific challenge

• Try new services in a sandbox using workloads that mirror your own

• Get direct answers on implementation constraints from AWS engineers and partners

Re-Invent runs November 30 to December 4 in Las Vegas.

Register now and start shaping your re:Invent agenda.

PRODUCTS

ChatGPT's meeting recorder has bigger implications

As a reporter and a generally type-A person, I have used AI notetakers for years. I currently rely on two, but ChatGPT's new feature may have just knocked one out of my rotation.

On Wednesday, OpenAI launched a Meetings plugin that functions as a note-taker in the newly launched ChatGPT Space tab. It works exactly as described: you hop into a meeting, it listens in, takes notes, generates a transcript, and then, at the end of the meeting, provides a detailed summary with bullet points and action items. 

I tried it with The Deep View's Jason Hiner on a quick catch-up call, and the summary and follow-up items were on the mark. The transcript also looked good in terms of transcription-to-word fidelity. However, there is a catch: the transcript does not have labels for who was in the conversation or options to add them, and you cannot go back and relisten to the audio. So it essentially functions as a Granola substitute rather than an Otter.ai replacement. 

As someone who uses Otter.ai and Granola for capturing meetings, there's a place for both. I like that using Granola feels less intrusive for the other people on the call, and, quite honestly, I rarely need an audio recording of my team meetings. Mostly, I want a rundown of what occurred and a way to quickly index what was said by asking the assistant, such as "When did he say the team dinner was?" That's something Granola and ChatGPT's new note-taking feature can do.   

While the version of Granola I use now is free, the incentive to switch to the ChatGPT version is that it helps me avoid context switching by keeping all my notes and transcripts alongside the chatbot I use every day for all my personal AI queries. In fact, OpenAI shares that everything ChatGPT knows about you and the work you have done together will be used to inform the personalized summary and next step section. That said, for calls in which I need to verify what was said and where accuracy matters, such as interviews for articles, I will have to keep using Otter.ai and its more thorough audio capture—and hefty subscription.

A winning move for OpenAI would be to concentrate more of AI's capabilities into its own application, as it would keep users from using third-party apps because of the convenience of having everything in one place and the added benefits that come from more ways to expand the context it has for a user, if they opt in. Of course, the dangers then become any one AI company wielding too much power by monopolizing the platform and hurting its competitors. Ideally, ChatGPT would become more of a hub where all the different applications live, with ChatGPT integrations building another layer. That's a vision Sam Altman says he's aligned with. At OpenAI DevDay, Altman told the media, "We want to figure out how to put this everywhere and drive people, revenue and customers and help people create these new things and make the most robust, richest ecosystem we can." 

LINKS

• US government, Meta, Google join Biohub in $1.8 billion AI bio effort

• Meta acted on 33.2 million child sexual exploitation items in H1 2026

• Healthleap raises $38 million for AI patient risk platform

• Musk says SpaceX will route tasks to best available third-party model

• Xreal Aura smartglasses open for pre-order at $1,279, with Gemini AI

• Anthropic cuts Sonnet 5.5 cache read prices in half

• GPT‑6: new GPT‑6 model built for more than 1.2 billion people who use ChatGPT each

• SynthID website: Google's new site that can identify AI-generated media 

• Google Playground: browser-based platform to make games using AI prompts 

• Claude Haiku 5.5: cheapest, fastest, and most capable small model released

• Smooth Media: AI & Automations Specialist

• Freight Hero: AI Deployment Accelerator

• Tata Consultancy Services: AI Automation Engineer

• Profitmind: AI Agentic Engineer

GAMES

A QUICK POLL BEFORE YOU GO