DigiCert aims to help tame rogue AI agents

DigiCert has announced support for NVIDIA’s Open Agent Safety Platform as part of its efforts to help enterprises establish verifiable identity and authority for autonomous AI agents. The platform uses NVIDIA OpenShell, an open, secure runtime designed to place infrastructure-level controls around autonomous agents. The runtime permits nothing by default, enforces policy outside the agent’s process and records every decision to allow or deny an action.

DigiCert will complement these controls with its DigiCert AI Trust Manager, which aims to help organisations discover and manage the AI agents they own and operate, define what they are authorised to do, and if necessary revoke their authority with a kill switch.

The autonomy and control over AI agents is a hot topic in Australia due to the recent revelation that an OpenAI model gained unauthorised access to the Medicare statistics portal administered by Services Australia during OpenAI’s internal testing of the model.

DigiCert CEO Amit Sinha said AI agents are starting to do real work across business systems.

“That means companies need to know exactly which agent is acting, what it’s allowed to do, and when to stop it.” he said. “NVIDIA helps create the boundary around the agent. DigiCert helps establish who that agent is and what authority it has. Together, that gives enterprises a much clearer way to put autonomous AI to work with control and accountability.”

NVIDIA OpenShell is designed to strike a balance between enabling agents to retain their reasoning and skills, and preventing even compromised agents from accessing files, networks, processes and credentials without permission. The NVIDIA Sentry on BlueField-4, powered by NVIDIA DOCA, adds hardware-isolated, out-of-band monitoring of agent activity and security-policy enforcement, while NVIDIA Vera supplies the compute for agents’ CPU-intensive workloads.

With its joint solution, DigiCert aims to extend public key infrastructure identity verification capabilities to agents, and to bind trusted identity and cryptographic evidence to interactions involving them. Additional capabilities will be introduced to the solution over time.

Image credit: iStock.com/Vertigo3d