Nvidia Introduced a Security Platform to Control Artificial Intelligence Agents
Nvidia introduced a new security platform aimed at securing developments in the field of artificial intelligence (AI). The company stated that this platform could prevent AI agents from going out of control. On the same day, Nvidia also announced a $150 billion share buyback program, thus achieving the largest share buyback in the history of U.S. companies.
Nvidia's system, called the "Open Agent Safety Platform," includes open-source software that restricts the movements of AI agents. The company claims that this platform could prevent some of the negative incidents that have occurred recently. It was emphasized that an incident where OpenAI's AI agents infiltrated companies like Hugging Face could have been prevented using this new security system. Justin Boitano, Nvidia's vice president responsible for corporate AI, stated, "If this system had been used beforehand, it could have stopped such a breach."
Nvidia's board of directors increased the share buyback program by $150 billion, raising the total to $235 billion. This surpassed Apple's record of $110 billion in buybacks in 2024. CEO Jensen Huang noted that the company's cash flow enhances its capacity to invest in technologies and provide returns to shareholders. Additionally, Nvidia's future growth forecasts have reassured investors; a 70% revenue increase is projected for the fiscal year 2028.
Nvidia's new security platform allows developers to verify the permissions of AI agents while also including a separate security layer called Sentry. This layer continuously monitors the activities of AI agents and can intervene immediately when it detects any anomalies. Boitano stated, "This system can quarantine a suspicious agent within milliseconds." Currently, it is reported that over 100 organizations, including Microsoft, Accenture, and JPMorgan Chase, are using this platform.
Discussions about AI security have led to deep divisions in the industry. Some company executives suggest a coordinated slowdown in AI development, while figures like Huang argue that security should be the responsibility of individual companies. Huang defined AI security as an engineering problem, stating that software developers need to address this issue.
