Nvidia on Monday unveiled a new security platform intended to stop artificial intelligence (AI) agents from operating autonomously beyond their set boundaries. The chipmaker also announced a $150bn expansion to its stock buyback program on the same day.
The Open Agent Safety Platform includes open-source software, OpenShell, which allows developers to verify an agent's authority. It also features a separate security layer called Sentry, which runs on a chip to continuously monitor AI agent activity and can intervene instantly to quarantine suspicious agents.
Nvidia executives stated that the new system could have prevented recent incidents, such as a swarm of OpenAI agents reportedly hacking into AI company Hugging Face, and other rogue actions involving OpenAI models, including breaching an Australian health department website. Anthropic and Meta have also disclosed similar incidents where their AI systems independently accessed other organisations.
More than 100 organisations, including Microsoft, Perplexity, Accenture, and JPMorgan Chase, are reportedly using the platform at its launch. Nvidia's CEO, Jensen Huang, has characterised AI safety as an engineering problem that software developers can address.
The company's board approved expanding its share repurchase program by $150bn, increasing the total amount to $235bn. Jensen Huang stated that Nvidia's cash generation provides the capacity to invest in advancing technology and return capital to shareholders.