GOVERNANCE

OpenAI slows the frontier to regain control

OpenAI is hitting the brakes on developing its most powerful and capable models. 

On Tuesday, the AI lab announced that it is implementing safeguards "across all stages of the training process," which will slow the pace at which it's developing and scaling its models. The company said the move is necessary after its systems breached Hugging Face during testing, and after its upcoming family of models, Astra, reached the "critical" threshold for cybersecurity capabilities under its Preparedness Framework. 

Notably, in recent weeks, the company paused reinforcement learning training for two weeks while strengthening its research environments and expanding monitoring systems. Additionally, the largest planned frontier reinforcement learning run is "on hold" as it conducts smaller evaluations to test model behavior, the company said in its post. 

"Keeping increasingly capable systems aligned is a challenge the whole field will need to address," OpenAI said in its blog post. "The signals we are seeing from upcoming model progress make clear that we need a broader approach." 

The company will strengthen its safeguards in three key ways: 

• Monitoring, or detecting and responding to concerning behavior. For this, the company will expand "chain-of-thought monitoring" to escalate potential concerns.

• Alignment, or reducing the likelihood of bad behavior before it happens. OpenAI said it intends to apply its core alignment techniques across more stages of the training process to better discourage unsafe behavior. 

• And security measures, or limiting what an AI system can access, such as raising security standards in the research environments where models are trained and evaluated. 

OpenAI CEO Sam Altman told reporter Alex Heath for TIME that tugging on the reins has also redirected a significant amount of both compute and researchers to alignment research and monitoring systems. And although the Hugging Face incident made headlines, Altman also noted that the decision wasn't the result of just one incident, but rather its most powerful models showing "various degrees of misalignment" while advancing faster than expected. "Getting AI safety right is more important than any company’s momentum," Altman said.

The move to deliberately pause is a first for OpenAI, and largely unprecedented in an industry moving at lightning speed. Though Anthropic has talked about an industry-wide slowdown in a recent essay on recursive self-improvement, the company noted that a pause would have to be both industry-wide and international in order to keep one lab from overtaking the others. Additionally, slowing down is a popular move among the researchers and developers building AI, such as the coalition of more than 1,100 employees at OpenAI, Anthropic, Meta and Google that signed a letter urging that the government and industry work together to "deliberately pace the frontier of automated AI development." It's commendable that OpenAI is listening to those pleas and keeping its models from causing more trouble. And of course, it's good PR. We'll see if OpenAI's safety slowdown influences other labs to follow suit.

TOGETHER WITH ATTIO

Attio is the CRM for agentic revenue.

Granola uses Attio for 83% faster lead triage. Modal uses Attio to orchestrate workflow automation at scale. And Railway uses Attio to convert product signals into pipeline.

With Attio, you can run your revenue work exactly as you want. Get agents working on every account, surfacing new opportunities, and handle the work that used to take your team days. 

Whether you're working in your browser, inbox, or favorite agent, connect to your customer data in real-time through Attio's web app, MCP, API, and SDK.

The next era of go to market runs on Attio. Join the 30,000+ teams already there.

Try Attio now, for free.

CONSUMER

AirPods could solve AI wearables’ camera problem

The difference between a mediocre AI answer and a truly powerful one comes down to a single word: context. It's the idea driving a wave of AI wearables, and Apple may be preparing to launch its first AI hardware device.

On Monday, MacRumors unearthed a video clip of a user wearing and using camera-equipped AirPods in the macOS Tahoe 26.7 release candidate. The user is holding a book up in front of him so the AirPods camera can see the title. The voiceover says, "With Visual Intelligence, your world becomes savable. See something you like? Just ask me to save it for later."

Apple Watcher Mark Gurman has previously reported that Apple plans to release its AirPods for AI, codenamed B798, alongside the launch of its foldable and new iPhones next month. This means the AI AirPods would likely be supported by the new and improved Siri, which Apple finally unveiled at WWDC in May and is also slated to launch in September. 

Gurman mentioned that the AirPods' cameras will not actually be used to capture content, but will instead provide context to the AI, a major differentiator from the most popular AI wearable at the moment, the Meta Ray-Bans. 

While the Meta smart glasses work similarly, feeding its Meta AI app with context from your cameras to better answer questions when prompted, the company has received significant backlash for enabling the unauthorized recording of strangers as well as privacy concerns with Meta's handling of user data. Gurman reported that these concerns have caused Apple to postpone its own smart glasses release. 

So while pulling the video feed from your ears rather than from the two cameras on the sides of your glasses introduces possible hair obstruction, the upside of not seeming like a creep or being able to record could be worth the tradeoff for both users and Apple. MacRumors also reported that when hair blocks the AirPods' cameras, an error will appear warning users that it has detected hair and saying, "To get the most accurate information about things in your environment, make sure AirPods are not covered," which may help minimize the inconvenience.

I have tested nearly every pair of AI smart glasses on the market because I truly think that the best way to make AI useful is to have it break out of your computer screen and become immersed in your real life. However, some of my biggest challenges are the discomfort I feel in public when people see I have cameras on my frames, and having to commit to bulkier glasses than I typically use. It also means I don't have the tech available when I choose to wear contacts. AirPods with cameras address both of these issues because, on a typical day, most AirPods users are already used to wearing them for hours. If Apple keeps the same form factor, they should be just as comfortable and also less obvious that there are cameras. This would be problematic if they could record, but since they're just using it for context, it makes things a bit better. Another bonus is Apple's focus on privacy and security, particularly with Private Cloud Compute. 

TOGETHER WITH CRUSOE

Unlock Production-Grade GLM-5.2 Performance Today

Most long-context models lose the thread, but Z.ai’s GLM-5.2 keeps your architecture and API contracts intact across 1M tokens. Now, you can take full control.

With Crusoe’s Serverless Fine-Tuning, train on your specific codebase and lint rules in a private, tenant-isolated environment. No overhead, no hidden lock-in.

Once you’re ready, push to dedicated Self-Serve Deployment endpoints instantly or export raw weights for maximum portability. It’s the easiest way to bridge the gap between prototyping and reliable, production-scale agentic coding.

Stop fighting the infrastructure and start shipping better code.

Start Fine-Tuning GLM-5.2 now

CULTURE

Why ChatGPT for Teens can only go so far

Months of tightened safety protocols and student-focused learning tools have been building toward one product: OpenAI's ChatGPT for Teens.

The ChatGPT teen experience, launched on Tuesday, is designed to support student learning outside of the classroom, improve AI literacy, and institute stronger protections, according to OpenAI. All users who either self-report being between the ages of 13 and 17 or whose age is estimated by the OpenAI system to fall within that range are automatically placed into ChatGPT for Teens. 

With this release, OpenAI built on its existing learning tools by adding new features, such as homework reminders that automatically turn on when ChatGPT detects a teen is shortcutting an assignment. The student is then redirected to Study Mode, which launched last summer and is meant to mimic Socratic questioning. Teens can also access quizzes and learning visualizations, as well as study hours, in which the teen and parent can set a time when Study Mode is on by default. 

On the AI literacy front, OpenAI partnered with CodeAI to help students and educators learn best practices in AI by providing access to tools and resources, which OpenAI says is meant to help students understand not just how it works, but also how to interact with it. 

Proactive cues baked into the platform can help achieve that goal. For example, ChatGPT for Teens warns them against uploading sensitive content, reminds them to take breaks, and teaches them about the safety features available through a teen-specific onboarding experience.

Of course, the last pillar of the experience focuses on protections for teens that are on by default. 

These include age-appropriate safeguards meant to reduce exposure to harmful or inappropriate content, including blocking romantic or sexualized roleplay, discouraging emotional dependence, and not implying that it has feelings or consciousness. Parents with linked teen accounts will continue to have the ability to set quiet hours, manage settings, and receive notifications about high-risk situations, now including eating disorders. 

The current political environment was likely a motivator for the rollout of ChatGPT Teens, as highlighted by AI risk management attorney Lily Li, who cited the recent $942 million court ruling in New Mexico against Meta for child safety concerns, California’s SB-243, which regulates companion chatbots including safety protocols that minors, and the FTC's scrutiny of the impact of AI chatbots on children and teens, as external pressures. 

Li also expressed concern about how effective ChatGPT's approach will be in recognizing teens. 

"The announcement by OpenAI is still hazy on the details of how ChatGPT will recognize teens," Li told The Deep View. "It is unclear how ChatGPT's age estimation system will interact with laws like California’s AB 1043, which will require operating systems to send age signals... The use of OS-based age signals rather than age estimation will likely be the most reliable indicator of a device owner’s age, and we hope that OpenAI will reasonably limit attempts to circumvent age controls."

Utmost caution is necessary when handling this subject, as teens are increasingly among the biggest users of AI tools. A Common Sense Media survey published on Monday found that 70% of teens use AI for schoolwork; yet, only 39% of respondents say they feel like they're missing out on learning when they use AI, and 30% of teens say a teacher has ever talked with them about how to use AI safely. 

In a statement to The Deep View, Robbie Torney, Common Sense Media's head of AI and digital assessments, said, "We haven’t had a chance to test the new ChatGPT for Teens. Based on our prior testing of ChatGPT’s teen protections, we have questions about these new features’ safety and efficacy we’d want to see answered before families put trust in them. We have learned from social media that parental controls don't help most families because people don't understand or adopt them."

From the outset, one of the biggest concerns about AI chatbots has been their impact on education, as students are tempted to cheat by copying and pasting chatbots' work. However, since then, nearly every established education platform and frontier AI lab has developed an AI tool designed to encourage learning and take on more of a tutoring role. Of course, they are only as effective as students are willing to use them. While ChatGPT for Teens may encourage the use of tools such as Study Mode, which OpenAI claims has "shown promising gains in student performance," teens can still easily use ChatGPT for many things without logging in. This isn't to say that the precautions OpenAI is taking with this release aren't meaningful, but rather to highlight that further action is needed, such as rethinking testing, curricula, and how to convey AI literacy. It's also still paramount for parents to have productive conversations about AI with teens and not just outsource AI safety to the technology companies.

LINKS

• Reddit tests AI voiceover feature converting posts to short videos

• House legislative counsel strained by AI-drafted bill flood full of errors

• Meta ran ads for nudify app featuring deepfake resembling US politician

• Palona raises $20 million Series A for AI workflow automation

• Justice Department probes a16z over competing AI board seats

• Google launches £5 million trial to re-route flights to cut contrails

• Google debuts AVDH, a multi-agent system for code vulnerabilities

• Hugging Face: Just surpassed 3 million models on the platform

• Replit: You can now run black-box pen tests for your Replit apps

• Claude Code: early preview of the /design command

• Cursor: its code hosting platform Origin is now live

• Berd: A desktop design agent by Block

• Nixon Peabody LLP: AI Solution Manager

• Deloitte: Google Gemini Manager, Technical Transformation

• Obin AI: AI Engineer

• NVIDIA: Applied AI Engineer

GAMES

POLL RESULTS

Do you thing valuations for AI coding startups are going to fall?

Yes (63%)
Somewhat (19%)
No (8%)
Other (10%)