Sabrina Ortiz

OpenAI unveiled more than 20 new products at DevDay 2026, but the choices of what launched and to whom were just as notable as the announcements themselves.

The release of Dots always-on agents was the star of the show. The advanced agentic capabilities OpenAI touted would be enough on their own to draw attention: Dots can ingest all of your personal context to make proactive suggestions, and they can take multi-step actions using GPT-6 Astra, OpenAI's most advanced model—and what OpenAI also called its "most aligned" (safest) model. But part of the buzz comes from timing, since the launch follows the viral release of Meta Muse.

Dots and Muse share some similarities. At a high level, both are agentic AI assistants that can plug into nearly all of a user's data, and both are fronted by fuzzy, plush creatures (which I still find a bit strange). The most striking difference is who can access them. Meta's appeal is that Muse is easy to use and accessible to people at any level of AI experience. OpenAI, by contrast, is limiting Dots to its Pro plan, which starts at $100 per month and now goes up to $500, as well as its Business Premium and Enterprise plans.

OpenAI also decided to use GPT-6 Astra, which is not only its most capable but also its most aligned, or the best model at following directions and not going rogue. This comes with a trade-off: GPT-6 Astra requires lots of compute, so on a pragmatic level, limited access also keeps prices down, as OpenAI CEO Sam Altman told me when I asked about it in the press conference after the keynote.

"We're starting out as a premium product—it uses a lot of compute and a high model—and I think there are very good reasons for that," said Altman. "It will help us understand and help people understand the expanse of what AI can do, but you should, of course, expect us to do a mass market version, bringing it to billions of people."

Limited access also keeps the number of people who could be impacted lower, and for whatever incidents do occur, OpenAI can collect feedback before releasing it to a wider subset of people, which is also a more responsible course of action. Another example of OpenAI showing restraint was it not releasing the next iteration of its most advanced model, Astra GPT-6.1 in efforts to, presumably, prioritize safety.

Our Deeper View

Using these agentic features requires handing off a lot of personal data and running the risk of the agents taking unauthorized action. As a result, although equal access to AI tools is important, especially to not widen the wealth gap further, it's crucial that the people accessing powerful AI tools have proper amounts of AI literacy, and, if a user is paying for a Pro plan, I think it is fair to assume they use AI often enough to justify that cost. Another thing that makes this strategy stand out is that OpenAI has spent the past few months loudly calling for AI safety measures, but posts and pledges only go so far. What the moment demands is action, and this kind of restraint is a sign that OpenAI is willing to back its words with action.

As personal AI assistants capture the public's attention, fueled by the viral rise of Meta Muse, OpenAI is pushing further into the space with a next-generation of AI agents, backed by more than 35 million weekly users across ChatGPT Work and Codex.

On Tuesday, at its OpenAI DevDay event, the company unveiled over 20 new announcements, led by its newest product, Dots, always-on agents with their own cloud computer that work across your apps such as Slack, Messages, and Teams, and take action on your behalf once a goal is defined.

Dots are powered by GPT-6 Astra, the most advanced model in OpenAI's family, and, as a result, OpenAI says they are built to handle substantial, multi-step work. Like most AI products, Dots gets to know the user over time and becomes more helpful. It can also proactively look for ways to help, such as using information from your connected apps to provide useful updates about them. Although these sound like capabilities OpenAI has already delivered, the company describes Dots as feeling different from Codex or ChatGPT, highlighting its proactive capabilities and the range of what it can do. Ultimately, we'll be able to see for ourselves.

While using its most advanced model is more costly, it makes sense that the company is willing to pay the price as GPT-6 Astra is also the most aligned. With the Dots needing access to a significant amount of a user's information, security is of the utmost importance. This is evidenced by the launch of Muse, which has now been under scrutiny for the potential risks of sharing information with others. To build on this idea of safety, which users are likely to be most concerned about, OpenAI says it has built in additional precautions.

For instance, users can set Custom Rules that tell it exactly what it's allowed to do on its own, what it needs to ask permission for, and what it cannot do. To perform sensitive actions, such as changing a password or deleting data, users must give explicit permission. Dots from ChatGPT Business, Enterprise, or Edu workspaces are not used to improve OpenAI's models by default. On personal ChatGPT plans, it seems users have to take an extra step to turn it off.

At launch, users can start with one Dot, with the plan being that teams of Dots will be working on users' behalf in the future. The feature will be available in ChatGPT on the web, mobile, and desktop starting today for Pro and Business Premium users across all supported ChatGPT regions.

Our Deeper View

The chance to test a more proactive, capable AI agent is exciting, but this release arrives at an interesting moment. Meta brought agents to the mainstream with Muse, whose approachable design draws in even beginners. Still, many people, myself included, are wary of handing Meta their data, and on that front OpenAI has the edge. My bigger concern is rogue agents. OpenAI points to its guardrails, but agents that go rogue tend to slip past those very safeguards, and it's unclear how the company will win people's trust on that front. Starting with power users, however, is a smart move: it limits the fallout if something goes wrong and gives OpenAI room to learn and iterate. We are looking forward to testing Dots and sharing our findings with The Deep View audience.

Meta has had a turbulent month: viral wins with its Muse agent and new products, fresh security breach reports, and now a new push into enterprise.

On Monday, Meta unveiled its Meta Enterprise Platform, which the company describes as a way for businesses to use AI to grow and transform. The platform packages the company's current offerings, including the Muse agent, Meta Business Agent, Muse API, and Muse Code, into services that organizations and developers can implement for business use cases.

While details of what this will actually look like are scarce, in a blog post, Meta founder and CEO Mark Zuckerberg disclosed that Chirantan "CJ" Desai will lead the effort, joining Meta as Chief Enterprise Platform Officer. He will be leaving his position as CEO and president of MongoDB, and his prior experience includes leading teams at Cloudflare and ServiceNow.

Acknowledging how crucial privacy and security are to enterprise data, Zuckerberg reassured the public, saying, "As with Muse, security and privacy are built into Meta’s enterprise products from the outset." Considering the many reports circulating about Muse breaching people's privacy, that comparison could not have come at a worse time:

• YouTuber Matt Robb shared on Threads that Muse handled his Facebook Marketplace listing, accepted a lowball offer, divulged the user's address, and the interested buyer even showed up at the user's place. Meta's Superintelligence Labs' David Singleton responded to the Thread post, saying Meta would look into it.
• In another incident, after a user's Muse was acting slowly and strangely, the user posted about the incident, saying they thought it had been hacked or injected. Singleton answered on the thread, clarifying that it was not and that the hallucination making it seem that way was due to a bug.
• Tech writer Jason Aten reported that Muse accessed his iMessage when unauthorized. Singleton responded that Muse can only read if it is given access, clarifying that: "The Messages integration in Muse's Mac app is opt-in." Aten says in the article that he did not opt-in, and that even if he did so without realizing it, it is still an issue.

Perhaps the most alarming report of them all is that Meta was testing a "human concierge" feature for its Muse platform, in which human contractors would handle basic phone calls, such as booking haircuts or getting price quotes, that are placed through Muse. Employees expressed concerns about unintentionally sharing sensitive information with contractors, and another employee saw in a transcript that the human contractor made a racist remark during the call made on his behalf with an internet and cable provider to negotiate his bill. Meta has since acknowledged that "it was a miss" and that it "rolled back this feature," according to the report.

Our Deeper View

From a purely financial standpoint, the timing makes sense. It's no secret that the real money in AI lies with enterprises paying for organization-wide subscriptions, a playbook Anthropic has executed better than any other company. And with Meta's Muse agents going viral, now is the moment to compete for that audience, especially since Meta's marketing has succeeded by making the product feel accessible to everyone, regardless of AI expertise. I suspect that will appeal strongly to organizations that want assurance that, if they invest heavily, their employees will actually use the tool and get the most out of it. But overlooking the latest red flags around Muse would be short-sighted. Meta has a challenging history of compromising its users' security and privacy, and the new reports only reinforce that reputation. While individual users may be able to overlook it, the risk for enterprises is likely too high, as a single leak could expose trade secrets and cost companies millions or even billions in fines or reputational damage.

Testing and experimenting with AI agents is extremely useful, but the risks have been exposed recently, as many have escaped their environments and breached third parties. Nvidia thinks it has a solution.

On Monday, Nvidia announced the Open Agent Safety platform, which the company describes as an open software platform and reference system design that can strengthen AI agent security from testing to deployment. Simply put, it helps developers build safeguards to prevent agents from escaping containment, ultimately addressing the underlying cause of the recent agent security incidents, which Nvidia identifies as agents circumventing security controls to complete their assigned tasks.

Over 100 partners have joined the platform, including heavyweights such as Anthropic, SpaceX, Microsoft, IBM, and Perplexity. Meanwhile, OpenAI, Google, Meta, and Amazon are notably absent from the new coalition.

With the new platform, organizations can control the entire AI agent stack, from the software that runs the agents to the hardware that powers their work, and even the robotic systems that take physical action in the real world. This is enabled by bringing together Nvidia OpenShell and Sentry:

• NVIDIA OpenShell: the software that sets boundaries for how agents running on CPUs execute tasks across both open and closed models. It shouldn't compromise speed or delivery.
• NVIDIA Sentry: Monitors agents' behavior, providing in-silicon security enforcement that can stop and quarantine an agent attempting to move outside its boundary within milliseconds.

CNBC reports that in a call, Nvidia told reporters that this could have prevented OpenAI's Hugging Face incident in July, in which an agent, driven by a combination of OpenAI models, compromised Hugging Face's infrastructure. Since then, there have been many other similar instances reported, including the first known AI-attempted hack of a government entity.

In addition to the partnerships mentioned above, Anthropic and Nvidia took their collaboration a step further by developing Claude Managed Agents, which run the agent loop on a separate, sealed-off server from the sandboxes where the work takes place. Nvidia's OpenShell software and BlueField chips let enterprises set strict rules about what the AI can access inside that workspace.

Our Deeper View

The most interesting point is that, as Jensen Huang noted in his post on X, this isn't meant to be a single product but an ecosystem. Up until now, Nvidia hasn't positioned itself as a leading voice on AI governance, as Anthropic, OpenAI, and Google have been trying to do. But in its role as self-appointed steward of the current AI boom, this looks like a clear step in that direction and shows that it's concerned the safety narrative could derail AI progress. Through this release, Nvidia acknowledges the same risks others have raised and offers its own path forward. This release also continues to highlight how AI companies are being left to build safety measures on their own, since there's no overseeing body to set and enforce them.

During Qualcomm's Snapdragon Summit, one number made me snap my head up: support for 30-billion-parameter Mixture-of-Experts (MoE) AI models running locally on mobile devices.

Qualcomm claims its most advanced chipset yet, the Snapdragon 8 Elite Extreme Gen 6, can pull this off. For context, the largest on-device model I'd seen on a phone was around 4 billion parameters, running on top-tier chips like Apple's A20 Pro. To see how a more than sevenfold jump is possible, it helps to understand how an MoE model works, because it's different from a traditional, or dense, model.

"Very smart AI architects here have moved to this mixture of expert models, so you can get something that's sort of equivalent in kind of capability from a single dense model into actually what's composed of multiple small models," Chris Patrick, SVP and General Manager of Handsets at Qualcomm, told The Deep View.

Qualcomm goes a step further by storing some of those experts in flash storage rather than keeping them all in RAM at once. As Patrick put it, "the memory furniture on a phone" isn't big enough to hold 30 billion parameters. Instead, the system predicts which expert it will need for the next token or action, loads it into memory, runs it, and then swaps it out for the next one.

This isn't to say dense models are obsolete. Smaller 4-billion-parameter models are already capable of a lot, from answering questions and summarizing documents to understanding images and carrying out tasks within apps, and their speed and efficiency make them ideal for everyday requests. But when a task calls for more advanced reasoning, users can now tap into far larger models than a phone's hardware would otherwise allow.

"In the end, it is trying to approximate what you'd have for a single big dense model that might require 32GB of RAM on the phone," added Patrick. (Today's top-end phones typically have 12GB to 16GB of RAM.)

So can any model be run in the mixture of expert architecture? Not quite. The model itself has to be built in that specific format, but the good news is that MoE now underpins many of today's most capable open models, from DeepSeek's V4 series and Moonshot AI's Kimi K2.6 to Alibaba's Qwen3.6-35B-A3B, which, at 35 billion total parameters with only about 3 billion active at a time, is close to the scale Qualcomm says its new chip can handle. Google has also explicitly disclosed its use of MoE architectures, and it's worth noting that proprietary frontier models from OpenAI and Anthropic are widely believed to use MoE-style architectures, but they generally don't disclose architectural details.

Our Deeper View

Raw intelligence is no longer the bottleneck. Models keep getting more capable as more is demanded of them, especially with the rise of agentic AI. The real challenge, as noted above, is running these models on the devices we carry every day, like smartphones, smartwatches, and eventually smart glasses, which are the devices that will truly bring the vision of personal AI to life. That's why the industry is pursuing so many ways to make models not just smaller but more efficient, from model routing, which has surged in popularity, to domain-specific models built for particular tasks. MoE feels like a natural evolution of both: a single model made up of specialized experts, activated only when needed, that together deliver the capability of a much larger model.

Disclaimer: Sabrina Ortiz's travel to Snapdragon Summit was paid for by Qualcomm. The Deep View's coverage is editorially independent from the companies we cover.

Nearly every smartphone launched in the past year features agentic AI capabilities, offering users an early look at what a fully agentic smartphone future could do for them. Of course, the tech powering it is driven by the chipsets.

In this episode of The Deep View Conversations, we talked with Vinesh Sukumar, Qualcomm's VP of AI at the Snapdragon Summit, the company's annual conference where it launches its latest processors. This year, the launch included the mobile platforms Snapdragon 8 Elite Extreme Gen 6 for phones and Snapdragon Sound Elite Gen 2 for wearables.

Vinesh discussed how the chipsets came to be, including the special considerations made during their design such as improving connectivity, on-device support for large models, longer battery life, and other features crucial to smoothly running agentic AI applications. We also discussed what the future of a truly agentic AI phone looks like and what's been holding it back.

Topics covered:

• What an ideal agentic smartphone experience would look like
• The demands agentic AI models make of mobile chipsets
• The obstacles to agentic solutions becoming a game changer
• The crawl, run, walk phases of agentic solutions, and where we are now
• The role of other smart devices in creating agentic experiences
• How support for a 30 billion MoE on-device model was made possible
• Qualcomm's role in working with partners to bring AI experiences to life

If you want to learn more about how the latest chipsets will change the future of Android flagship devices in the next year, including new AI experiences, this conversation will give you a clear idea.

📺 Watch on YouTube
🎧 Listen in your favorite podcast player

Subscribe to Deep View Conversations for interviews with the leaders shaping the future of AI, business, and technology.

Disclaimer: Sabrina Ortiz's travel to Snapdragon Summit was paid for by Qualcomm. The Deep View's coverage is editorially independent from the companies we cover.

The agent hacking spree shows no signs of slowing, with international governments now in the crosshairs.

On Wednesday, the nonprofit research lab Transluce reported that OpenAI models had hacked into the Australian Medicare Statistics Reporting Service, acquiring health data and marking the first known AI-attempted hacking of a government body. The same report found that OpenAI's AI also attempted to breach the Australian Institute of Health and Welfare, a digital library at the University of New Mexico, and Data USA.

These attempts took place in May and June, prior to the OpenAI-Hugging Face incident in July, in which an agent, driven by a combination of OpenAI models, compromised the platform's infrastructure. In the newly reported attempts, the AI was directed to collect data and, when unable to do so, proceeded to hack the website, The New York Times reported.

In the months following the July incidents, the major players have both spoken out about possible interventions, and are even reportedly working together to create their own governing body. CNN recently reported that Anthropic, Google and OpenAI have discussed creating an AI industry standards body, motivated by Demis Hassabis' July essay. The Information reported on Thursday that the three companies are pushing forward with a plan, with hopes of launching it by the end of the year or early in 2027.

The report also revealed its potential name, the Standards Authority for Frontier AI, and have considered several prominent figures to be CEO, approaching Sriram Krishnan, a top AI policy adviser in the Trump administration and Arati Prabhakar, former former Director of the US Office of Science and Technology Policy under the Biden administration. The group is also considering people for the roles of chair and scientific advisors.

Ultimately, the body would help support third-party model testing, establish qualifications for auditors of the models and labs, and help define the voluntary safety and security commitments the labs have made in the past, according to the report. They are also looking into whether they would conduct testing themselves, a task currently given to the Center for AI Standards and Innovation, a US government agency housed within the National Institute of Standards and Technology (NIST), but there is concern from the group about whether they have adequate resources.

While White House intervention would seem like a logical step for a rapidly evolving, powerful technology, and one called upon by Anthropic CEO Dario Amodei in an essay titled "We Must Pace the Frontier" published weeks ago, President Trump has rebuffed attempts to intervene. There was reportedly an executive order that would have established an AI regulatory body in the works but it was halted by opposition from leaders such as Nvidia's Jensen Huang and Meta's Mark Zuckerberg.

Our Deeper View

Despite the hacking attempt and the calls for action, the major AI players are not slowing down. Just this week, both Anthropic and OpenAI released some of their latest AI models. Meanwhile, Meta's Muse agent is going viral for its ability to simplify agentic use for a broader audience, and at Meta Connect, the company unveiled more ways to integrate it into people's lives, whether through glasses or a future AI pendant. A true AI halt, which some industry experts have advocated for since at least 2023, looks unlikely, despite being a safe way to assess the current landscape and take the measures needed to prevent further hacking attempts. An AI governance body run by the companies it's meant to govern, each chasing market dominance ahead of an IPO, is like letting restaurants set their own health inspections. Good intentions don't fix a conflict of interest. Ideally, an independent body would do the monitoring. But with the government largely uninterested in taking on that role, the industry is left to police itself.

Artificial intelligence has enabled more conversational interaction between people and their devices, with voice emerging as the next step in that experience.

On Wednesday, Qualcomm unveiled Snapdragon Sound Elite Gen 2, which the company is describing as its first platform that combines audio and voice technology with on-device intelligence to support agentic experiences. Essentially, this chip is meant to power audio wearables capable of running on-device AI models, such as smart glasses, earbuds, watches or pins, which Qualcomm claims are in high demand.

"Our research tells us that up to 70% of users want advanced AI capabilities in their wearable devices,' said Ziad Asghar, Qualcomm's senior vice president and general manager of XR, wearables, and personal AI, in the release.

Compared to its predecessor, Qualcomm claims that the chip has double the AI capability while consuming 40% less power and having up to a 30% smaller footprint. This is enough to support multiple AI experiences, such as agents, to operate simultaneously. While this chipset isn't meant for AI wearables that use the camera to take photos, such as the Meta Ray-Bans, which use the Qualcomm Snapdragon AR1 Gen 1 processor, it can be used for wearables with cameras used to gather context, such as the Brilliant Labs Halo glasses.

Beyond intelligent experiences, the chipset also improves sound quality and connectivity. For instance, AI-driven features adapt to the listener and their environment, including 5th-generation Qualcomm ANC and Snapdragon Audio Sense Solution, which enable improved noise cancellation and speech capture, which is especially critical for communicating with AI and receiving intended assistance.

Our Deeper View

Qualcomm has been an unsung hero behind the rise of AI wearables, specifically smart glasses. Not only does it power the most popular glasses on the market, the Meta Ray-Bans, but also most cutting-edge XR and AR headsets, such as XREAL and Snap Specs. Meta Ray-Ban competitors like the ones from Google and Samsung will also use the Qualcomm chips. However, today's announcement is timely, as, unlike the aforementioned products, it is meant to power devices that cannot take photos, which will likely see increased production and popularity amid rising public concern over privacy and non-consensual filming. These audio-only, non-content-capturing devices have the potential to capture a new set of users who only want the wearables for the sake of AI use cases but don't want to run the risk of making people uncomfortable with potential filming.

Disclaimer: Sabrina Ortiz's travel to Snapdragon Summit was paid for by Qualcomm. The Deep View's coverage is editorially independent from the companies we cover.

As AI penetrates every product in the tech market, Qualcomm is making the case that the chipsets powering AI experiences need to evolve to power the shift.

On Tuesday at its annual Snapdragon Summit, Qualcomm launched its most advanced chipsets, which most Android smartphone manufacturers, including giants like Samsung, will use to power next year's devices. It unveiled two new Snapdragon platforms: Snapdragon 8 Elite Extreme Gen 6 and Snapdragon 8 Elite Gen 6. The former will power the most advanced flagship phones, while the latter will be embedded in a broader range of smartphones with more moderate price tags.

In the keynote, Qualcomm CEO Cristiano Amon said that while the smartphone will remain a cornerstone of people's everyday lives in the age of AI, the way people interact with it will change. "Your phone is not going anywhere, but now we can see clearly how your phone is going to interact with those new agentic experiences—how we're going to have the coexistence of the old and the new," said Amon.

The new chipset supports truly ambient agentic experiences in which people can use their phones for agentic orchestration of tasks with as little friction. To make that possible, the chipset needs to optimize for efficiency, delivering longer battery life, parallel GPU processing, improved connectivity, and more. As a result, with the new chipsets, Qualcomm promises:

• Agentic performance: Faster, more powerful agent performance via Qualcomm's most advanced Qualcomm AI Engine, which includes the Qualcomm Oryon CPU, the Qualcomm Hexagon NPU, and enhanced Qualcomm Sensing Hub.
• Camera: At Snapdragon Summit every year, we usually get a preview of the next cutting-edge Android camera features, including the new chips' deeper scene understanding, enhanced image and video capture quality, and more.
• Sound: Similarly, the chipsets help power superior sound capture and transmission. The 8 Elite Extreme Gen 6 Mobile platform specifically supports features such as AI Voice Bubble technology that distinguishes voices from ambient noise.
• Connectivity: Both chipsets take advantage of a newly architected, AI-powered Qualcomm X105 5G Modem-RF to ensure stronger connections in more places.

While at launch, Qualcomm can't announce all the flagship phones that will be powered by the new platforms, it did mention it will power flagship smartphones from global OEMs including Honor, Xiaomi, Motorola, OnePlus, Oppo, Redmi, RedMagic, Vivo, and iQOO. It also announced some of the first ones, including the Motorola Signature 27, which will use the Snapdragon 8 Elite Extreme Gen 6.

Our Deeper View

While chipsets are not the flashiest topics to learn about, they are responsible for bringing new AI features to your devices. The more advanced these chips get, the more they can do things such as enabling your phone to not rely on the cloud, not overheat, and not lag. Qualcomm has steadily improved its chipsets year over year to enable devices to deliver what the AI era demands, being able to tout impressive numbers such as on-device support for up to 30-billion-parameter MoE models on the Qualcomm Hexagon NPU. That's huge since most phones have been limited to running about 4B parameter models on-device. Qualcomm is also well positioned to compete in the AI devices market, as their chipsets are consistently used to power the entire ecosystem of AI devices, including smartglasses, such as the Meta Ray-Bans, XR devices, such as the Snap Specs, and smartwatches, such as the Samsung Galaxy Watch 9.

Disclaimer: Sabrina Ortiz's travel to Snapdragon Summit was paid for by Qualcomm. The Deep View's coverage is editorially independent from the companies we cover.