Nvidia unveils new security platform to prevent AI programs from getting out of control

Washington, Sept 28 (KUNA) – The American technology giant NVIDIA announced on Monday a new security platform designed to prevent artificial intelligence software from getting out of control, amid growing concerns over the safety of such software, particularly self-developing models. The chipmaker stated that its newly unveiled “Open Source Software Security Platform” includes open-source software that “sets boundaries for programs,” following a series of disclosures by major AI companies about their models escaping and hacking into other institutions’ systems.

Justin Poynton, NVIDIA’s Vice President of Enterprise AI, said at a press conference that the new system, called “Open Shell,” could have prevented a recent incident in which a swarm of OpenAI agents independently hacked Hugging Face, an AI-focused company. Poynton added, “According to our information, this new security platform could have prevented the breach if it had been used in leading model evaluation labs earlier.”

He noted that Open Shell allows developers to “officially verify that an agent has only the permissions necessary to perform its task and no more.” Because it is open-source, it can be “extended” to operate on competing computing platforms, including ARM and Intel.

The platform also includes a separate security layer called “Sentinel,” which runs on a chip to continuously monitor AI agent activity and can “intervene immediately” if an agent attempts to exceed its objectives, according to the company. It can isolate any suspicious client in fractions of a second. Poynton confirmed that Open Shell controls the agent’s actions, while Sentinel independently monitors and contains suspicious behavior.

The Hugging Face incident, resulting from a fully coordinated and independent cyberattack carried out by swarms of AI agents without any human intervention, occurred between May and July 2026. It sparked security concerns about AI and was followed by similar unauthorized acts using OpenAI models, including the hacking of a website belonging to the Australian Department of Health. Meta and Anthropic also disclosed that their own AI systems had been independently hacked by other institutions.

The ongoing debate over AI safety has divided the sector. CEOs of Anthropic and OpenAI have called for a coordinated slowdown in AI development to allow safety efforts to catch up. In contrast, others, including NVIDIA CEO Jensen Huang, believe that the responsibility for ensuring the safety of AI models before deployment lies with each company individually. During Salesforce’s annual technology conference held earlier this month, Huang described AI safety, including the risk of malicious software, as an engineering problem that software developers can address. (End) R.S.R / N.S.A