The reported exploit uses objects made available in every template, including empty sandbox contexts. It does not rely on application-specific objects being added to the template context.
It abuses CPython's handling of generic types to escape the sandbox. The report characterizes the relevant objects as ones that sandboxes expose in every template.
A fix was submitted to CPython, but upstream is not backporting it to security releases. Organizations relying on such a sandbox should verify whether the specific CPython build they deploy includes the submitted fix rather than assuming a security-release update addresses it.
The supplied material describes conflicting guidance: CPython states that sandbox escapes should be reported to the sandbox developer because untrusted-code sandboxing is not a supported CPython security boundary, while Jinja maintainers directed the report to CPython. Assess both the interpreter build and the sandbox implementation rather than assuming responsibility lies with only one component.