NVIDIA Brev security breach may have exposed SSH keys used to access customer VMs

NVIDIA has disclosed a security incident on its Brev cloud development platform that may have exposed SSH keys used to access customer virtual machines. Unauthorized activity was reported between October 4 and October 20, and the company rotated affected keys and updated customer instances with replacement keys. NVIDIA said it found no evidence of improper use, but it has not explained how attackers gained access or whether customer data or workloads were accessed.

What NVIDIA said

According to a notification sent to affected customers, some accounts may have been exposed. NVIDIA said it investigated the incident and identified unidentified activity on the platform, adding that information stored on the platform may have been exposed to malicious actors. The affected information included the SSH key set up by Brev. "We are writing to inform you that NVIDIA Corporation has experienced a security incident that may have affected the security of your data," the notification reads. The company said the affected SSH key had since been rotated and customer instances updated with the replacement key. It did not explain how attackers gained access, or whether anyone attempted to use the compromised data, and it did not disclose whether customer data or workloads were accessed.

Why SSH keys matter

SSH is a protocol used to establish a secure connection between virtual machines or remote computers working together. Developers or system administrators can use it to manage remote servers, transfer files, and execute commands without needing physical access to the machine. Brev uses SSH keys to help customers access these services in the cloud. If attackers access these keys, they could potentially take control of a virtual machine, depending on its specific permissions and other security controls.

Broader security context

The incident comes as several firms hosting AI development services and cloud infrastructure grapple with mounting security incidents involving customer credentials. On October 30, cybersecurity firm Wiz released a blog called "Black October" for AI developer platforms. The firm identified at least nine platforms that had security incidents during the month, including Zeabur, Fireworks AI, Lightning AI, Opencode, Warp, and Northflank, among others. Wiz said the incidents were recurrent, with researchers noting that "all public reports highlight targeted exploitation, explicitly hunting stored environment variables, with a focus on those granted AI access." Wiz also shared two cases involving compromised Amazon Bedrock credentials: in one, attackers generated about 800,000 API calls over 30 hours, while another key was used to make roughly 1.7 million calls over 13 hours. The combined usage came to $100,000 at list prices.

Other incidents and recommended checks

Such incidents are not restricted to NVIDIA. Major US banks including JP Morgan, Morgan Stanley, and Bank of America restricted agents from accessing customer accounts earlier this month, mainly because automated assistants could veer off set instructions and initiate unauthorized approvals. Bradford Regeski, a threat intelligence analyst at RH-ISAC, recommended extensive checks for customers in SSH-related incidents, including isolating hosts that installed the compromised extension during the exposure window, removing unauthorized SSH keys and persistence mechanisms, and rotating credentials and sessions. Updating compromised software may not be enough to eliminate an attacker's access.

Sources: Tom's Hardware