ManageIQ CVE-2026-52903

Requires an authenticated ManageIQ (or CloudForms) account that holds dialog import privileges - not an anonymous or read-only user. The instance must be running in production mode, because the vulnerable YamlLoadAliases fallback from YAML.safe_load to YAML.unsafe_load is specifically the production behavior. The attacker must supply a YAML payload containing a class disallowed by safe_load so that Psych::DisallowedClass is raised, which is the exact event that activates the unsafe deserialization path. Limiting factors: the dialog-import authorization requirement (PR:L) confines this to insiders or already-compromised accounts, and CloudForms being EOL means affected operators cannot patch and must rely on restricting the dialog-import role.

This is a credible, high-priority vulnerability for any live ManageIQ deployment, but with important caveats. The CVSS 3.1 vector (AV:N/AC:L/PR:L/UI:N) correctly reflects network-reachable, low-complexity exploitation requiring only low-level authentication, and the C:H/I:H/A:H impact matches a full remote-code-execution outcome via Ruby object deserialization (CWE-502) - a well-understood and reliably weaponizable class of bug. The design flaw is especially concerning because it is not an edge case: the YamlLoadAliases module deliberately falls back from YAML.safe_load to YAML.unsafe_load in production whenever a Psych::DisallowedClass error occurs, meaning a disallowed (malicious) class is precisely the trigger that activates the unsafe path. No EPSS score, KEV listing, or public POC was provided in the input, so there is no public exploit identified at time of analysis and no evidence of in-the-wild activity - the real-world threat is currently potential rather than demonstrated. The main mitigating factor is the PR:L requirement: the attacker must already hold an authenticated account with dialog import access, which bounds this to insiders or post-compromise lateral movement rather than anonymous internet scanning. Note also a scope nuance: Red Hat CloudForms (one of the two affected CPEs) is End of Life and will receive no patch, so CloudForms operators carry unremediable residual risk and should treat this as compensating-control-only.

An authenticated ManageIQ user who has been granted dialog import permission uploads a maliciously crafted YAML dialog definition. The payload references a Ruby class disallowed by safe_load, which raises Psych::DisallowedClass and silently triggers the production fallback to unsafe_load, deserializing the attacker's gadget chain and executing arbitrary code on the appliance. No public exploit was identified in the provided data, but deserialization-to-RCE via Ruby/Psych is a well-established and reliable technique given the required privilege.