The vulnerability allows an attacker who can supply an expression to the default asteval Interpreter with numpy enabled to access a ctypes type exposed by the library. By using the ctypes.from_address function and an unfiltered attribute assignment, the attacker can obtain an arbitrary memory address via id and then read or write that memory location. This memory read or write can corrupt process memory, execute arbitrary code, disclose secrets, or crash the host process, effectively escaping the intended sandboxed environment.

The flaw affects the lmfit:asteval product, specifically any installation of asteval older than version 1.0.9. The issue is present in the default Interpreter implementation before the 1.0.9 release.

The CVSS score of 5.5 classifies the vulnerability as moderate. EPSS data is not available, and the issue is not listed in the CISA KEV catalog, suggesting that widespread automated exploitation is currently not expected. The likely attack vector is an expression supplied by an untrusted source to the asteval Interpreter; once such an expression is evaluated, the attacker can perform the memory operations described. The mitigation is straightforward through an official patch, but an unpatched system remains susceptible to privilege escalation or data compromise within the process.