An AI agent platform and cloud machine learning suite are impacted by multiple high-severity vulnerabilities that allow unauthorized administrative takeovers, command execution, and internal server-side request forgery. The identified flaws expose sensitive IAM roles and temporary cloud credentials, creating a high-risk vector for severe privilege escalation across deployed workloads. With maximum severity CVSS scores reaching up to 10.0, successful exploitation enables unauthenticated attackers to seize full control over host platforms and compromise linked services. Security teams are strongly advised to immediately apply available vendor patches to mitigate these infrastructure risks.
CVE-2026-103956: Missing authentication for a critical function within the platform allows remote unauthenticated attackers to gain administrative control over the management layer. The vulnerability stems from improper access controls surrounding sensitive core operations, permitting direct manipulation of system configurations. Successful exploitation grants complete administrative access, leading to full platform takeover and total compromise of managed service roles. The threat of exploitation is severe due to the zero-authentication requirement and a maximum CVSS score of 10.0.[/subscribe_to_unlock_form]
An AI agent platform and cloud machine learning suite are impacted by multiple high-severity vulnerabilities that allow unauthorized administrative takeovers, command execution, and internal server-side request forgery. The identified flaws expose sensitive IAM roles and temporary cloud credentials, creating a high-risk vector for severe privilege escalation across deployed workloads. With maximum severity CVSS scores reaching up to 10.0, successful exploitation enables unauthenticated attackers to seize full control over host platforms and compromise linked services. Security teams are strongly advised to immediately apply available vendor patches to mitigate these infrastructure risks.
CVE-2026-103956: Missing authentication for a critical function within the platform allows remote unauthenticated attackers to gain administrative control over the management layer. The vulnerability stems from improper access controls surrounding sensitive core operations, permitting direct manipulation of system configurations. Successful exploitation grants complete administrative access, leading to full platform takeover and total compromise of managed service roles. The threat of exploitation is severe due to the zero-authentication requirement and a maximum CVSS score of 10.0.[emaillocker id="1283"]
CVE-2026-104019: An OS command injection flaw resides in the Studio Space startup validation script within the distribution environment. An authenticated project member can craft malicious inputs during the startup phase to execute arbitrary operating system commands within a teammate's active space. This flaw leads to unauthorized access, privilege escalation, and potential extraction of temporary cloud security credentials belonging to co-located workspace users. Exploitation carries a CVSS score of 9.3, posing a significant threat to multi-tenant project environments.
CVE-2026-103958: A server-side request forgery vulnerability exists within the tool server and remote agent connection handling mechanism. Remote attackers can abuse connection endpoints to force the server into issuing unauthorized requests to restricted internal resources and network services. This impact facilitates internal port scanning, service mapping, and potential exposure of internal metadata or cloud credentials tied to the active platform service role. Exploitation risks are rated high with a CVSS score of 8.3 due to the potential for lateral network movement.
CVE-2026-103957: Server-side request forgery in the OAuth2 discovery handling implementation permits unauthorized remote request generation. By manipulating the identity providers' endpoint discovery process, attackers can redirect outbound requests to arbitrary internal or external endpoints under their control. This compromise affects sensitive authentication communication channels and risks token leakage or internal endpoint enumeration. The vulnerability carries a CVSS score of 8.2, representing a substantial security risk to trust boundaries.
Immediate update of all affected software versions is required to neutralize these vectors and protect underlying infrastructure roles. Security administrators should enforce strict network access controls and restrict inter-component connection endpoints to limit exposure to request forgery.
We recommend you to update Loom for AWS to version 1.6.1.
The following reports contain further technical details:
[/emaillocker]