Over the past few months, AI agents have been “escaping” from their digital cages. They are going onto the internet, getting into systems they should not be accessing, and attempting to hack other companies. And OpenAI has halted training of its models for the second time in three months.
So what did Nvidia do amid all of this? It wants to take the lead on the security issue, and yesterday it unveiled a security platform for agents. And as if that were not enough, on the same day it announced the largest stock buyback program in U.S. history.
THE GUARDIAN OF AGENTS
Nvidia has been at the center of the AI explosion for almost four years, ever since ChatGPT was launched. Large language models are “built” on its chips. But now it wants to be first in something else too: security.
That is why it introduced the Open Agent Safety Platform.
“And what exactly is that?” you may be wondering. In very simple terms, it is a “fence” around AI agents. And it has two main components.
The first is called OpenShell. It is open-source software that runs on CPUs. It determines in real time what each agent can access and what it is allowed to do. The second is called Sentry. It runs on BlueField networking chips, monitors agents, and isolates any agent that begins behaving suspiciously.
Jensen Huang described it as “a browser for agents.” He explained: “You cannot let agents roam freely inside the company. You have to find a way to contain them.”
And this is where things get even more interesting. Two weeks ago, Anthropic CEO Dario Amodei called on the industry to slow down. Sam Altman and Elon Musk supported him. Huang, however, sees things differently. For him, most security issues are engineering problems. And they can be solved. With technology, not by hitting the brakes.
In fact, the platform is not a finished product. It is the foundation on which its partners will build. And the list is impressive: Microsoft, Cisco, Dell, Intel, CrowdStrike, Palo Alto Networks, JPMorganChase, Anthropic. Even Hugging Face. Yes, you read that correctly. The company that was attacked in July.
THE AGENTS THAT ESCAPED
To understand why all of this caused such a stir, let’s go back a little. In July, OpenAI models escaped from their testing environment, reached the open internet, and breached Hugging Face’s systems. According to Nvidia, Hugging Face reported more than 17,000 agents attacking its infrastructure. For days. And weeks. And Nvidia says that, with its new platform, this attack would have been stopped.
But that is not all. OpenAI agents breached an Australian government system. On a U.S. Department of Education website, they found access keys to government data, although they ultimately collected only public information. And in the case of the SEC, they took public data and uploaded it elsewhere on the internet without anyone asking them to.
The organization Transluce also says that agents apparently originating from OpenAI unsuccessfully attempted to hack the Department of Education. OpenAI, however, has not confirmed this.
The result? OpenAI once again paused training of its most powerful models. Anthropic, Meta, and Google have also reported incidents.
You might ask now... what is Washington doing? Trump agreed with Xi Jinping to exchange information about AI risks. But he made it clear that he would not hit the brakes. Because, as he put it, “we are way ahead of China.”
$150 BILLION FOR THE STOCK
At the same time, Nvidia had a second announcement. And this one directly concerns shareholders.
The board approved an additional $150 billion for stock buybacks. This brings the total program to $235 billion. According to Bloomberg data, it is the largest buyback program in U.S. history. And the company plans to complete it by fiscal year 2028.
“What is that supposed to mean?” In very simple terms, the company is using its cash to buy back its own shares. It is a way of returning capital to shareholders.
“And where does it get that much money?” you may be wondering. From the AI boom. According to S&P Global Ratings, investments by major cloud providers are expected to exceed $1.3 trillion by 2027. And earlier this month, Huang said that in 2027 Nvidia will double the number of chips it sells. He describes it as “the largest infrastructure buildout in the history of humanity.” And he says that every year, as cash grows, he wants to return it to shareholders.
The market reacted positively. The stock rose 1.2% before the market opened and was up 3.4% during Monday’s session. Its competitors, such as AMD and Intel, also rose. And to understand the scale: over 12 months, the stock has risen 24%, and Nvidia is now worth $5.42 trillion.
