A week in, and it’s already doing something

Nvidia doesn’t do things halfway. Just one week after the Open Secure AI Alliance (OSAA) launched, the group has already formed a working group with a name that’s hard to forget: the Shared AI Findings Exchange, or SAFE. It’s already publishing proposals for public comment, with the Linux Foundation — itself a member — managing the process.

The group put this together while its members were gathered at Black Hat, the massive cybersecurity conference happening this week in Las Vegas. That timing wasn’t accidental. The people who care about AI security were already in one room, so why not get to work?

What the proposals actually cover

Let’s be honest: the initial proposals aren’t going to blow your mind. They’re practical, not flashy. The focus is on three things:

• How to confidentially report AI cybersecurity incidents
• How to alert the people who might be affected
• How to do blame-free analysis so everyone can learn from what happened

That last one matters more than it sounds. In cybersecurity, the instinct is often to point fingers. A blame-free approach means companies actually share what went wrong, instead of hiding it. That’s how the whole industry gets smarter.

More than just talk: actual tech contributions

The OSAA isn’t just writing documents. Members are also contributing open source tools that could eventually become a real, usable stack for securing AI agents — or defending against rogue AI attackers.

Consider what’s already on the table. Nvidia has a family of open models and an open source LLM vulnerability scanner called Garak. Okta is working on agent identity tech. Red Hat is focused on agent governance. Amazon brought both an open agent-building tool, Strands Agents, and an authorization language called Cedar. And that’s just the start.

This is the kind of concrete, technical work that makes an industry group worth paying attention to. Anyone can sign a letter. Actually contributing code is another thing entirely.

Who’s in, and who’s conspicuously absent

The membership list is impressive: Adobe, BlackRock, Cisco, Intel, Microsoft, and Visa are all in. That’s over 120 companies total.

But look closer and you’ll notice some big names missing. Anthropic, OpenAI, and Google are not members. That’s notable, because both OpenAI and Google signed the original open letter that led to this group’s creation. The letter, published last week, urged the White House to support open source AI rather than suppress it. Nvidia championed it, and over 200 tech companies signed on.

Anthropic’s absence isn’t surprising — they’ve been skeptical of open source AI for a while. But Google? They’re generally seen as a big open source supporter, and they’ve released open weight models of their own. So why aren’t they in?

It’s possible they’re waiting to see how the group evolves. It’s also possible they’re just slow to commit. Either way, it’s worth watching whether they join as the Open Secure AI Alliance builds momentum.

Moving at AI speed

The speed here is remarkable. It’s only been a couple of weeks since reports surfaced that the Trump administration might ban Chinese open weight models. That caused real consternation in the industry, which led to the open letter. And now, a week after the OSAA formed, it’s already publishing proposals.

That’s not normal for industry groups. Most take months just to agree on a mission statement. This one is already doing work.

What this means for open AI in the US

Whether or not Chinese models get banned, the fast action from this heavyweight group looks like a good thing for the U.S. open AI ecosystem. Some in the ecosystem, like the co-founder and CTO of U.S. open weight AI lab Arcee, argue that openness is the best way to counter any threat — real or imagined — from Chinese AI labs.

“Openness may be one of the most important paths to AI safety and security,” the group wrote in their letter. That’s a strong statement. The good news is they’re backing it up with action, not just words.

For anyone tracking AI security trends or open source AI developments, this is a group worth keeping an eye on. It’s young, it’s fast, and it has the backing of some of the biggest names in tech. What it does next will tell us a lot about where open AI is headed.

Seoul’s grand AI experiment

South Korea is about to do something no other major economy has tried. The government wants to hand every citizen a free AI chatbot with zero usage caps. No tiers. No premium plans. No “ask your IT department.” Just a national assistant that works like a public utility.

The Ministry of Science and ICT announced the AI for Everyone project on July 13. Private companies will build the platform around locally developed models, while a separate AI agent will help people navigate government services. It’s a more practical job than generating emails or settling arguments nobody wanted to research themselves.

What will the free chatbot offer?

The plan covers two distinct services. First, a general-purpose chatbot that anyone in South Korea can use without paying. Second, a public-service agent that identifies relevant government programs and helps users complete applications. Think of it as a digital civil servant that never sleeps.

Two or three private operators will be selected to develop these services. A beta is expected by the end of September, with an official launch planned before the end of 2026. The government will supply up to 512 Nvidia B200 GPUs, although the chosen companies must also invest their own capital.

Why must the AI stay local?

Here’s the catch: at least half of each service must run on South Korean foundation models that meet the ministry’s standards. Developers using their own models must source more than 30% of the system from other domestic AI companies. Foreign alternatives can fill limited gaps, but the government won’t subsidize them.

The rules keep more public funding inside South Korea’s technology industry while reducing its reliance on overseas platforms. A national service isn’t especially dependable if a foreign provider can suddenly tighten its limits or cut off access. That’s a lesson many governments are learning the hard way.

What about the tech giants?

This isn’t just about sovereignty. It’s about competition. South Korea is home to Samsung, LG, and a vibrant startup scene, yet its citizens largely rely on American chatbots. The government wants to change that by funding a domestic alternative that’s good enough to pull people away from established commercial platforms.

Can free AI remain free?

Government support is scheduled to continue through the end of 2030. Its scale from 2027 onward will depend on annual evaluations and budget discussions, so the longer-term definition of “free” isn’t entirely settled. Applicants have until August 11 to submit proposals.

The economics are tricky. Running a large language model for an entire population costs real money — in electricity, in GPUs, in cooling. The B200 GPUs are powerful, but they’re not infinite. If usage surges beyond projections, the government might need to rethink the model or set fair-use policies.

What this means for the rest of the world

South Korea is moving beyond research grants and limited trials by funding AI access for an entire population. That’s a bold statement about how a society should treat artificial intelligence — as infrastructure, not as a luxury subscription.

The harder question is whether its domestic models will be good enough to compete with the likes of OpenAI‘s ChatGPT or Google‘s Gemini. Local models have improved dramatically, but they still lag in some benchmarks. The September beta should give everyone the first useful answer.

If it works, expect other countries to follow. If it fails, expect the opposite — a cautionary tale about the limits of state-funded AI. Either way, South Korea is forcing the conversation.

For now, the clock is ticking. Proposals are due in August, the beta lands in September, and the world will be watching. Will free AI become a citizen’s right? South Korea is betting it can.

The rumor mill had it wrong

For months, the chatter around OpenAI’s first hardware product pointed to something flashy — a wearable, maybe even the seed of a future smartphone. Turns out, the company’s debut gadget might be far more mundane on the surface, yet far more personal underneath. According to a new Bloomberg report, OpenAI is building a smart speaker.

Not just any speaker, though. The device is being designed as a humanlike AI companion that plugs directly into your smart home, helps control appliances, plays media, answers questions, and taps into the full range of OpenAI‘s ChatGPT capabilities. People familiar with the matter say the defining feature won’t be sound quality or screen size. It’ll be personality.

More companion than speaker

Bloomberg’s sources say OpenAI internally views this not as a speaker at all, but as the company’s first AI-native computer. Think of it as the physical embodiment of ChatGPT — a presence that stays nearby and proactively helps throughout the day, rather than sitting idle until you bark a command.

The hardware will reportedly include a mechanical design with directional flexibility. That means it can move, tilt, or orient itself toward you, making it feel a little more alive than the static speakers we’re used to. It’s a concept that echoes what Apple has reportedly been tinkering with for the HomePod — a smart display on a robotic arm — but OpenAI is taking a screen-free route, focusing entirely on voice and AI interaction.

Eyes and ears

At the heart of the interaction will be GPT-Live, ChatGPT’s voice-first mode designed for natural, flowing conversation. The device is also said to include a camera for understanding its surroundings, plus an environmental sensor that could detect context — similar to Amazon’s presence-sensing Echo devices. The goal is an assistant that doesn’t just hear words but comprehends the room around it.

A battery means it follows you

One practical detail stands out: the speaker will have a rechargeable battery. That means it won’t be tethered to a wall outlet like every other smart speaker on the market. Instead, you could carry it from the kitchen to the bedroom, letting the AI companion stay close all day.

Bloomberg adds that this is just the beginning. OpenAI is reportedly working on five different hardware products, with the speaker expected to be the first to ship in 2027. The broader effort is being shaped by LoveFrom, the design studio founded by legendary former Apple design chief Jony Ive, who has been collaborating with OpenAI CEO Sam Altman on this new family of devices.

Timing raises eyebrows

The timing couldn’t be more awkward. Apple recently filed a lawsuit accusing OpenAI of poaching employees and using them to obtain confidential hardware information. Yet if Bloomberg’s reporting is accurate, OpenAI’s first device isn’t trying to be a smartphone killer or a display replacement.

Instead, the company appears to be betting that the next big AI gadget is something far more personal — a companion that quietly follows you through your day, understands your surroundings, and feels less like a gadget and more like someone always ready to help. Whether that vision lands by 2027 remains to be seen. But for a company known for software, this is a bold first step into the physical world.

For more on how AI is reshaping everyday tech, check out our take on AI-powered smart home devices and ChatGPT’s evolving voice features.