QWED-AI vulnerabilities and threats

CVE-2026-55585 is a high-severity code injection vulnerability in the QWED-AI qwed-verification package prior to version 5.1.2. The vulnerability arises because user-supplied math expressions are passed directly to SymPy's parse_expr() function without restricting the global and local namespaces, allowing arbitrary Python code execution. The API endpoints /verify/math and /verify/batch accept these expressions with minimal normalization and no sanitization. An attacker with any valid tenant API key, which can be obtained via the default-enabled signup and API key issuance endpoints, can exploit this flaw to execute arbitrary code, read or write files, run OS commands, terminate the service, and compromise other tenants in shared deployments. This issue is fixed in version 5.1.2.

CVE-2026-55546 is a critical code injection vulnerability in QWED-AI's qwed-mcp prior to version 0.2.1. The function verify_math_expression() improperly passes attacker-controlled input to SymPy's parse_expr() without restricting built-ins or validating the expression, allowing arbitrary code execution. Exploitation requires a downstream integration that invokes this function with untrusted input. The issue is fixed in version 0.2.1.