A recent report by SecurityWeek highlights that major companies, including Microsoft, Intel, Google, Walmart, and Amazon, are at risk of having their artificial intelligence (AI) infrastructure servers taken over due to flaws in the TorchServe open-source package. Researchers from Oligo have identified two critical vulnerabilities, known as CVE-2023-43654 and CVE-2022-1471, affecting tens of thousands of TorchServe instances. These vulnerabilities could enable remote code execution and remote unauthenticated access, allowing attackers to compromise the systems. The researchers also noted a default misconfiguration issue that further exacerbates the situation.
The exploitation of these vulnerabilities can lead to the complete hijacking of the AI infrastructure servers. Attackers can access and modify sensitive data flowing through the TorchServe server, potentially causing harm to the credibility and trust of the applications that rely on the AI infrastructure. Interestingly, the flaws may not require lateral movement, indicating that attackers can directly exploit the model serving server without the need for additional compromise steps.
To mitigate these risks, Oligo researchers recommend immediately updating to the latest version of TorchServe released in August, which includes fixes for these vulnerabilities. By promptly applying the update, organizations can protect their AI infrastructure servers from potential exploitation.
The impact of these vulnerabilities is significant given that they affect major companies using TorchServe for their AI infrastructure. This emphasizes the importance of regularly patching and updating open-source packages to ensure the security and integrity of critical systems.